Modulating autophagy to treat diseases: A revisited review on in silico methods

Graphical abstract

Highlights

• •Autophagy is a conserved evolutionarily pathway to maintain cellular homeostasis for human health.
• •Databases containing autophagy have been constructed to store information on genes, proteins, compounds, and diseases.
• •The systems biology approaches involve integrating multiple omics data to predict potential molecular interactions and linking autophagy with other biological processes.
• •The integration of the omics methods and artificial intelligence can build models and extract features.
• •Mathematical modeling combined with deep neural networks is used to describe the complex dynamic process of autophagy.

Abstract

Background

Autophagy refers to the conserved cellular catabolic process relevant to lysosome activity and plays a vital role in maintaining the dynamic equilibrium of intracellular matter by degrading harmful and abnormally accumulated cellular components. Accumulating evidence has recently revealed that dysregulation of autophagy by genetic and exogenous interventions may disrupt cellular homeostasis in human diseases. In silico approaches as powerful aids to experiments have also been extensively reported to play their critical roles in the storage, prediction, and analysis of massive amounts of experimental data. Thus, modulating autophagy to treat diseases by in silico methods would be anticipated.

Aim of review

Here, we focus on summarizing the updated in silico approaches including databases, systems biology network approaches, omics-based analyses, mathematical models, and artificial intelligence (AI) methods that sought to modulate autophagy for potential therapeutic purposes, which will provide a new insight into more promising therapeutic strategies.

Key scientific concepts of review

Autophagy-related databases are the data basis of the in silico method, storing a large amount of information about DNA, RNA, proteins, small molecules and diseases. The systems biology approach is a method to systematically study the interrelationships among biological processes including autophagy from a macroscopic perspective. Omics-based analyses are based on high-throughput data to analyze gene expression at different levels of biological processes involving autophagy. mathematical models are visualization methods to describe the dynamic process of autophagy, and its accuracy is related to the selection of parameters. AI methods use big data related to autophagy to predict autophagy targets, design targeted small molecules, and classify diverse human diseases for potential therapeutic applications.

Introduction

Autophagy, which is an evolutionarily conserved lysosomal degradation pathway, sequesters unneeded cellular cargo into an autophagosome with a double membrane, degrades abnormal cytoplasmic components as well as upgrades intracellular substances to maintain cellular, tissue, and organismal homeostasis [1], [2]. Recently, numerous studies have shown that mutations in autophagy genes leading to dysregulation in pathophysiology are associated with human diseases [3]. Accumulating studies of autophagy mainly focus on deciphering the intricate molecular mechanisms and crucial signaling pathways, revealing the complicated interrelationships between autophagy and human diseases. Ultimately, candidate drugs targeting autophagy can be developed to treat a variety of human diseases, including cancer, cardiovascular, neurodegenerative, immunity, metabolic, gastrointestinal, pulmonary and musculoskeletal disorders [4], [5], [6], [7], [8], [9], [10], [11]. However, the underlying mechanisms for regulating autophagy in human diseases have not yet been adequately elucidated. At first, it has been thought that the underlying regulatory mechanism of autophagy is post-transcriptional regulation. Subsquently, the transcriptional network has been reported to play a role in this crucial process. Some nuclear receptor proteins have been also reported to regulate autophagy by controlling the expression of autophagy genes [12]. A better understanding of the relationship between autophagy and nuclear receptors facilitates deciphering the autophagy systems. Developing specific compounds to induce and inhibit autophagy is a promising precision treatment strategy with many good clinical outcomes [13]. Apatinib targets VEGFR2/STAT3 signaling, significantly triggering autophagic cell death in lung cancer [14]. Targeting the activation of TRPV1 channels enhances autophagy to clear alpha-synuclein and boosts the therapy of Parkinson's disease (PD) [15]. G protein-coupled receptor kinase 4 mediates HDAC4 phosphorylation and inhibits Beclin-1 expression to decrease autophagy, leading to aggravation of cardiomyocyte injury [16]. Designing or screening small-molecule drugs targeting G protein-coupled receptor kinase 4 to inhibit its activity can enhance the treatment of ischemic heart injury. Targeting autophagy to treat diseases is inseparable from an in-depth understanding of the relationship between autophagy and diseases for the high accuracy of targeted selection. However, off-target effects of the small-molecule compounds may weaken the efficacy and even lead to adverse side effects [17]. Experimental methods can verify the conjecture and reveal the function of autophagy, but it is difficult to systematically describe autophagy and come up with a new idea by integrating huge and complex data. In the face of massive experimental data hiding important biological laws, in silico methods can be utilized to analyze big data and interpret the relationship between autophagy modulating and diseases at the genetic level to provide an insight into precision therapies of autophagy-related diseases (Fig. 1). Of note, there are still many mysteries in autophagy research, the main root of which is the difficulty of monitoring autophagy and the molecular mechanisms of autophagy and diseases [18]. The experimental method of studying autophagic flux has applicable results at the cellular level, but it is difficult to apply the to the physiological level and carry out in clinical trials. With such extensive autophagy-related data, autophagy research relying only on the current experimental methods seems to be limited. In silico approaches have been utilized to capture, compile, annotate and analyze the autophagy-related big data for conferring possible therapeutic potential in comparison with the experimental data [19]. For instance, Gene Ontology (GO) [20] has been reported to be the earliest database for studying autophagy and contains a large amount of autophagy-related information; thereby providing a basis for subsequent enrichment analysis and offering a stepping stone to drug development. The widely used method is omics, especially transcriptomics, which relies on RNA sequencing to analyze differentially expressed genes and further construct key protein interaction networks for subsequent research on the regulation of autophagy. Artificial intelligence (AI), the most popular in silico method, performs with a powerful performance in predicting protein structures [21]. Targeting autophagy from the molecular level to achieve precision medicine in silico approaches can assist experimental methods for extracting an insight into revealing autophagy mechanisms, optimizing drug scheduling, and treating diseases. We summarized the current in silico methods, including major databases (Fig. 2A), biology network approaches (Fig. 2B), omics-based analyses (Fig. 2C), mathematical models (Fig. 2D), and AI technologies (Fig. 2E), to aid in analysis of large data sets in the search for potential therapeutic implications to modulate autophagy.

Fig. 1.

The autophagy-in silico approach-human disease network.

Fig. 2.

The flowchart of in silico approaches for modulating autophagy. A. Autophagy-related databases. Collect and collate autophagy-related RNAs, proteins, compounds and diseases to construct multiple types of autophagy databases for storing and analyzing datasets. B. An example of systems biology approaches for research on autophagy. Modulize gene similarity networks using Gene Ontology Annotation to analyze the expression of core autophagy-related genes in autophagy networks about other biological processes. C. Genomic analysis to search for mutations in structural and functional genes associated with autophagy to provide gene targets for diagnosis and therapy. Bioinformatics analyses of transcriptomic data involve differential gene expression analysis, gene function annotation analysis and core autophagy-related protein interaction network construction to study autophagy at the transcriptional level. Proteomic analyses examine differences in autophagy-related gene expression at the protein level in given cells to aid in anchoring drug targets. D. Taking the metabolic changes of three nutrients as an example, the relationship between autophagy and diseases is discussed from the perspective of metabolism. E. Mathematical models for describing autophagy. For example, after cleaning the expression data of autophagy-related genes, suitable parameters are found based on differential equations and the local dynamic process of autophagy is modeled by algorithms. F. Artificial intelligence approaches for regulated autophagy: autophagy-related data are applied to artificial intelligence model training.

Autophagy and human diseases

Depending on how cargo is transported to the lysosome, autophagy is divided into three categories, macroautophagy, microautophagy and chaperone-mediated autophagy. Macroautophagy (hereafter called autophagy) is common to deal with environmental and physiological triggers to maintain cellular trophic homeostasis and normal organelle function. In addition to non-selective somatic degradation, selective autophagy, including mitochondria, lysosomes, peroxisomes, ribosomes, proteasomes, endoplasmic reticulum, nuclei and lipid droplets, plays a vital role in regulating human diseases [22], [23]. Moreover, non-canonical autophagy adds complexity to the molecular mechanisms of autophagy and opens up new possibilities for the treatment of diseases associated with canonical autophagy defects. Autophagy is a double-edged sword, with both promoting and inhibiting effects on normal organisms and tumors [24], [25]. On one hand, autophagy can enhance the tolerance of tissue cells or tumor cells to stress and maintain their survival in adverse environments, but on the other, autophagy can inhibit cell production and metastasis at different stages of cell development, and even serve as a death pathway for some apoptosis-deficient tumor cells. Autophagy inhibition can weaken the resistance of cells to the pressure of cancer drug treatment and promote the death of cancer cells [26]. However, autophagy activation can promote the release of ATP by tumor cells and the recruitment of immune factors to enhance tumor-specific immune responses to inhibit tumor growth and metastasis [27], [28], [29]. CircCDYL, an autophagy-related circRNA, promotes breast cancer progression through the miR-1275-ATG7/ULK1-autophagy axis augmenting autophagy [30]. Reduced SRSF1 enhances autophagy by decreasing the interaction with PI3K and partially blocking Bcl-x splicing to inhibit lung cancer progression [31]. Autophagy activation via circCUL2 suppresses tumorigenesis and ameliorates gastric cancer patient outcomes [32]. Weakening autophagy initiation via lysosome acidification of doxorubicin can reduce cardiotoxicity [33]. Autophagy dysfunction reduces the quality control of proteins and organelles in neurons resulting in neurodegenerative disorders [34]. PD is a representative progressive neurodegenerative disease resulting from the misfolding and aggregation of monomeric α-synuclein. Autophagy dysregulation bloks α-synuclein species to degradation in the lysosomes and abnormal mitophagy fails to clear damaged mitochondria well [35]. Selective absence of neuronal autophagy due to ATG5 loss promotes excitatory nerve transmission and leads to neuronal cell death [36]. Activation of autophagy seems to be beneficial for preserving the homeostasis of neurons and managing neurodegenerative disorders. Alterations in autophagy have participated in the etiology of cardiovascular conditions because positively disposing of multiple autophagic substrates is vital for maintaining cardiovascular fitness. Genetic and pharmacological inhibition of autophagy promotes disorder progression in several animal models relevant to cardiovascular disease [37]. Deficiency of ATG16L1, a crucial adaptor of mature autophagosomes, contributes to diminishing basal autophagy and triggering inflammatory signaling for generating Crohn's disease [38]. Due to the double functions of autophagy, there are two ideas for the clinical treatment of targeting autophagy: inhibiting protective autophagy and promoting autophagy-dependent death of cancer cells for improving the therapeutic effect in cancer, inhibiting excessive autophagy and boosting protective autophagy to therapeutic efficacy in neurodegenerative diseases or heart failure. The complexity of autophagy is also reflected in its extensive connection to other biological processes, such as normalizing components of autophagy can be key nodes in other signaling pathways. Studies have shown that impairment of LC3-associated phagocytosis affected by autophagy proteins in the myeloid compartment activates the regulation of tumor growth by tumor-related macrophages to increase tumor immune tolerance [39]. With unusual molecular mechanisms involving reactive oxygen species (ROS) stress, mitochondrial dysfunction, endoplasmic reticulum stress, and calcium ion concentration control, cell death pathways determine the fate of cellular protein included in human disorders by triggering autophagic pathways for responding to changes in the intracellular and extracellular environment [40], [41]. Computational methods in the face of complex autophagy mechanisms have unique properties that can help carry out experimental research more systematically. Thus, we discuss the therapeutic potential of using the in silico approaches to explore new promising strategies targeting autophagy to treat human diseases.

Databases for autophagy modulators

Hitherto, several autophagy-related databases have been generated focusing on RNAs, proteins, compounds, and diseases related to autophagy. More than being autophagy repositories, databases have built-in analysis tools, such as molecular docking, target prediction, and Functional enrichment analysis to analyze, study and visualize primary data for further guiding the research of modulating autophagy to treat diseases. In this section, we briefly introduce databases involving autophagy with updated states and carry out database resources in the form of comparison (Table 1).

Table 1.

Comparisons of the autophagy-related databases.

Database	Protein	Compound	RNA	Analysis tool	Mechanism	Disease	Target prediction	Weblink	Latest update	ref
HAdb	234	+	+	+	–	–	–	http://www.autophagy.lu/	May. 2022	[42]
GAMdb	197	–	836	–	+	56	+	http://gamdb.liu-lab.com/index.php	Jan. 2016	[44]
AutomiRdb	90	–	493	–	–	18	–	http://www.chen-lab.com/index.php	May. 2015	[45]
HAMdb	796	841	132	–	+	+	+	http://hamdb.scbdd.com	Jul. 20I8	[47]
NcRDeathdb	–	–	121	+	+	–	–	http://www.rna-society.org/ncrdeathdb/	May. 2015	[48]
Autophagydb	2163	–	–	–	+	–	–	http://tp-apg.genes.nig.ac.jp/autophagy	Oct. 2010	[50]
THANATOS	191,543	–	–	+	+	–	+	http://thanatos.biocuckoo.org/	May. 2017	[49]
AutophagySMdb	71	∼10,000	–	+	+	+	+	http://www.autophagysmdb.org	Jan. 2019	[51]
ACdb	–	357	–	–	+	+	+	http://www.acdbliulab.com/	Sept. 2017	[52]
ATdb	–	–	658	+	–	+	–	http://www.bigzju.com/ATdb/#/	Jan. 2020	[68]

Note: “+” indicates that the database contains the contents of the header item, and “–” indicates that the database does not contain the contents of the header item.

To investigate autophagy-related gene regulation in various types of diseases, the human autophagy database (HAdb) was originally constructed to aid in studying autophagy to treat breast cancer [42]. The researchers constructed HAdb in conjunction with an oligonucleotide chip to analyze the differences in gene expression of autophagy gene processes to interpret autophagy interactions to find drug resistance mechanisms. HAdb with evolution provides new capabilities involving analysis and visualization to facilitate prediction and exploration. Some research supported that microRNAs (miRNAs) played vital roles in autophagy pathways, such as mir223 targeting ATG16L1 inhibited autophagy to aggravate central nervous system inflammation [43]. Containing 197 protein targets and 836 miRNAs that regulate human autophagy, the gerontology-autophagic-miRNA database (GAMdb) bridges miRNAs and neurodegenerative diseases with autophagy to elucidate mechanisms of gerontology [44]. The autophagy-miRNA database (AutomiRbd) is an integrated database involving 493 miRNAs, 90 targeted proteins, and 18 cancers related to autophagy. AutomiRbd covers the complicated relationships between miRNA-associated autophagic mechanisms and cancers at the systems level to facilitate an understanding of regulating autophagy networks to promote cancer-related therapy in the future [45]. The Autophagy Regulatory Network is a comprehensive online resource containing transcription factors, miRNAs and protein–protein interactions within autophagy [46]. Based upon mining literature manually, ARN obtains 386 miRNAs, 413 transcription factors and 4,013 protein–protein interactions from 1,485 nodes. To reveal the complexity between autophagic pathways and autophagy-related diseases, the human autophagy (HAMdb) modulator database provides 796 proteins, 132 miRNAs and 841 chemical compounds to facilitate users in obtaining information on autophagy-related diseases [47]. Novel noncoding RNAs (ncRNAs) involving long noncoding RNAs and small nucleolar RNAs are continuously verified with the crucial function of regulating autophagy. However, relationships between upstream and downstream keep unclear in global signal transduction to impede drug development targeting ncRNAs. The ncRDeath database, a specialized repository for ncRNA-associated autophagy defects linked with human disorders, highlights the organization of ncRNA-mediated programmed cell death interactions [48]. The Autophagy, Necrosis, ApopTosis OrchestratorS (THANATOS) is a large integrated bioinformatics resource of 191,543 proteins and a variety of posttranslational modifications for studying autophagy in 164 eukaryotes [49]. In addition to basic information about autophagy-related protein data and their homolog data from 41 other species, the autophagy database provides structural and functional information on proteins and explores potential functional homologs from other species [50]. The diversity of targets provides a wealth of options for small molecule intervention. To facilitate researchers to search and understand the molecular mechanism of small molecule intervention, some databases of autophagic small molecule regulators have been constructed. The autophagy small molecule database (AutophagySMdb) covering approximately 10,000 small-molecule compounds and 71 proteins in human diseases provides various computational tools and data for users to identify scaffolds for designing candidate drugs and discover new therapeutic targets with experimental validity [51]. The autophagic compound database (ACdb), a novel library of autophagic compounds, contains multiple pieces of information about compounds, potential targets, pathways and related diseases for a better understanding of the interrelationship between the autophagy process and diseases [52]. Autophagy plays different roles in the progression and metastasis of cancers in diverse environments. As a comprehensive online resource connecting 25 tumors with 137 genes relevant to autophagy pathways, the autophagy and tumor database (ATdb) provides a platform to bridge tumor and autophagy data, extending a series of analysis tools (i.e., gene-related survival analysis) and visualizing alternative splicing in cancer [53]. Thus, the system data decentralized to establish multiple databases is conducive to quickly completing the database construction process, but there is also a disadvantage, even though database resource sharing, will also cause users to access inconvenient because of the self-rules of each database. Autophagy Regulatory Network seems to bridge this gap by integrating and unifying the multiple autophagy resources into one database. However, this unification only facilitates the user to consult the autophagy-related literature directory and does not provide all the information on keywords from multiple databases at one time. This problem can be solved by establishing a unified nomenclature and autophagy database establishment rules. In the above content, it is not difficult to find that we have not introduced full data resources, because many databases have a long history but have not been updated, and in the context of hot research on autophagy, the stagnation of databases will lead to a decrease in the reliability of data.

Beyond the above-specialized autophagy databases, there are other well-explored comprehensive databases covering autophagy. By using “autophagy” as the index word in the gene expression omnibus database, 840 results that involved next-generation sequencing and high-throughput microarrays were found [54]. Based on single-cell RNA-seq and multi-omics experimental data, array expression plays a vital role in mining the function of relevant genes [55]. More information regarding autophagy-related genes and products can be obtained from Gene Ontology resources [20]. Information on autophagy is also available in the Kyoto Encyclopedia of Genes and Genomes (KEGG) and the Reactome Knowledgebase [56], [57]. KEGG, the encyclopedia of genes and genomes, assigns genes’ biological functional significance at the molecular level for practical application in diseases and drug design. Manually curated molecular details and human biological processes can be derived from the Reactome Knowledgebase, which plays a vital role in the analysis of high-throughput data. As indispensable stepping stones for other methods, various types of databases relevant to autophagy covering autophagic biomarkers, regulatory mechanisms, linkages with other biological processes, and disease information are resources for modulating autophagy.

Systems biology approaches for modulating autophagy

The biological network is a complex network in which nodes can be genes, RNA, proteins, metabolites, etc. The edges of the network correspond to the physical, biochemical and functional interactions between the nodes. The interaction between biomolecules is dynamic because the edges between nodes can change in response to changes in internal gene regulation and the external environment. Network analysis is the focus of biological networks. The significant changes of biomolecules and their interactions in the network form a differential network, which has crucial reference value for cell signaling, cell development, environmental stress, disease state transition and drug treatment [58]. Systems biology approaches are applied to holistically study the composition of all the components (genes, RNAs, proteins, etc.) in a biological system and the interrelations among these components under certain conditions. The emerging therapeutic strategies targeting diverse physiological processes and pathological conditions related to autophagy, rather than individual factors such as genes or proteins, focus on holistic analyses that provide intermodular connectivity to promote the formulation of therapeutic strategies. In this section, systems biology methods are listed to conduct further experiments to decipher autophagic mechanisms and evaluate the crosstalk between autophagy and autophagy-related diseases (Table 2).

Table 2.

Systems biology approaches for modulating autophagy.

Title	Disease	Method	Feature	Application	Ref
A systems pharmacology approach to identify the autophagy-inducing effects of Traditional Persian medicinal plants.	Diabetes and cancer	Systems pharmacology approach, enrichment analysis	The autophagy-inducing effects of Traditional Persian Medicine are identified.	Development of TPM-based prescriptions	[61]
Autophagy-dependent survival is controlled with a unique regulatory network upon various cellular stress events.	Neurodegenerative diseases and metabolic diseases	Systems biology techniques	Kinetic features are explored. The process in which autophagy inducers turn on autophagy in a cell stress-specific manner was demonstrated.	The mechanism of the control network	[62]
Model-based pathway enrichment analysis applied to the TGF-beta regulation of autophagy in autism.	Autism	Network analysis, enrichment analysis	The differential expression of distinct biological pathways is detected. System-level model-driven approaches are applied.	To differentiate between disease conditions	[63]
Solamargine induces hepatocellular carcinoma cell apoptosis and autophagy via inhibiting LIF/miR-192-5p/CYR61/Akt signaling pathways and eliciting immunostimulatory tumor microenvironment.	Hepatocellular carcinoma	Weighted correlation network analysis	The eigengenes associated with the Solamargine are clustered. Activation of autophagy by Solamargine via multiple signaling pathways has therapeutic effects on hepatocellular carcinoma.	To identify differentially co-expressed gene modules	[64]
Network analysis of the progranulin-deficient mouse brain proteome reveals pathogenic mechanisms shared in human frontotemporal dementia caused by GRN mutations.	Frontotemporal dementia	Weighted correlation network analysis	Pathogenic mechanisms are revealed. 26 modules of highly co-expressed proteins are identified.	The development of targeted effective therapies	[65]
An Integrative Systems Biology and Experimental Approach Identifies Convergence of Epithelial Plasticity, Metabolism, and Autophagy to Promote Chemoresistance.	Lung cancer	Phylogenetic clustering, systems biology analyses, and molecular experimentation	Evolutionary convergences. Revealing vulnerabilities for treating therapy-resistant cancer.	The application of combining evolutionary and systems biology methods	[67]

Systems biology approaches for autophagic mechanisms

To explore the mechanisms of autophagic regulation at the molecular, cellular and tissue levels, an integration strategy of computational methods and network analysis has been used to track autophagy processes. A network analysis approach was applied to reveal that autophagy network adaptations to lipid-limiting conditions can be provoked by RAB18 loss, during which phosphorylation-dependent ATG9A is activated to modulate basal autophagic activity [59]. Furthermore, the outside-in functional signaling network has been identified in matrix-regulated autophagy, characterized by a paradigmatic shift for signaling cascades that maintain tissue homeostasis to combat autophagy-related diseases [60]. Systems biology techniques also aid in the observation of dynamic features during autophagy induction. A new systems pharmacology approach has been reported for identifying the functions of traditional Persian medicinal plants in autophagy, in which network centrality analysis, docking, and molecular dynamics simulation power the development of synergistic therapies for type 2 diabetes [61].To illustrate autophagy-dependent survival, a unique regulatory network with feedback loops has been utilized to study switch-like characteristics of the kinetic properties during the abnormal autophagy process in diseases [62]. By using systems biology methods, researchers have transformed the description of biochemical communication within autophagy into more in-depth studies on the biomolecular level and function, and deepened the discussion and understanding of the nature of autophagy. It is possible to consider autophagy as a relatively independent, functional 'small system' within the whole organism, but the complexity of this small system should not be underestimated, and with the current state of the art and level of analysis, a comprehensive and correct analysis of autophagy is not yet possible. With the emergence of new technologies and instruments, new data such as new high-throughput histological data are expected to emerge, and new analytical methods will be established to analyze the data and exhaustively annotate the information in the database to break the bottleneck in the development of systems biology.

Systems biology approaches for autophagy-related diseases

Network analysis methods have been implemented in interpreting autophagy-related data to decipher the association of autophagy with relevant diseases and reveal the organizing principles and potential drug targets for both the diagnosis and cure of disorders. Network analysis based on a system-level dynamic model contains enrichment analyses, such as pathway enrichment analysis that helps to classify autophagy-related (ATG) genes into relevant signaling pathways and extend the detection range of methods to further distinguish the health and disease conditions of the patients. This new model-driven approach exploits computer modeling to help researchers more intuitively understand the dynamics of molecular systems in diseases with extensive reference significance [63]. Recently, a weighted correlation network analysis was executed based on the transcriptomics sequencing dataset to identify the eigengenes of the modules relevant to solamargine. When combined with multiple analysis methods, the mechanism of solamargine as an autophagy inducer in the treatment of hepatocellular carcinoma was revealed [64]. A novel systems biology strategy of weighted correlation network analysis was performed, identifying that three modules of co-expressed proteins correlate to granulin gene (GRN) deficiency and insufficiency of progranulin defects in autophagy, thereby providing a new avenue for revealing the pathogenic mechanisms of human frontotemporal dementia related to GRN [65]. Furthermore, changing the signal transduction network caused the characteristic metabolism of cancer cells in which vulnerability was driven by mTOR, which inhibits autophagy, leading to an energy crisis and promoting drug resistance in cancer cells [66]. A conformable systems biology strategy combined phylogenetic clustering and molecular experimentation of epithelial-mesenchymal transition to reveal TGF-β-mediated chemoresistance relevant to autophagy and new therapeutic fragility for lung cancer [67]. Moreover, target prioritization relying on network methods has accelerated precision therapy. Based upon the random forest algorithm, a creative strategy has been developed into finding new autophagy-disease-related genes (ADGs), users grasped the association between autophagy and autophagy-related diseases from a system perspective [68]. Then, network pharmacology, based on systems biology theory, combined with molecular docking technology elucidated the mechanism of drug influence on autophagy and completed a multitarget drug design for the treatment of muscle injury. With the application of advanced tools such as machine learning and deep neural networks, more and more tumor susceptibility genes, including autophagy-related genes, are being identified and their location in the human genome, the types of cancers they affect and the association with sex-differentiated expression are being analyzed from a biosystematic perspective, promising new opportunities for personalized cancer prevention [69]. The systems biology approach makes up for the deficiency of local modeling and provides a comprehensive understanding of the autophagy system at the macro level. A world effect is shown by network analysis, which has guided a significance for multitargeted therapy of diseases.

Omics approaches for modulating autophagy

Omics methods in this article refer to the means of performing high-throughput sequencing and analyzing omics data. According to the different analysis goals, it is mainly divided into genomics, transcriptomics, proteomics, and metabolomics. Omics methods combined with bioinformatics can be used to analyze the differential levels of autophagy regulators, reveal new mechanisms of autophagy regulation, and elucidate the relationship between autophagy and other key biological processes. In this section, various omics-based analyses are summarized to identify targets that regulate autophagy for curing diseases (Table 3).

Table 3.

Multi-omics analyses for modulating autophagy.

Title	Disease	Method	Feature	Application	Ref
Quantitative proteomic analysis of temporal lysosomal proteome and the impact of the KFERQ-like motif and LAMP2A in lysosomal targeting.	Cancer	Proteomic analysis, multiplex quantitative mass spectrometry approach	LAMP2A on KFERQ-like motifs and lysosomal targeting are described.	Other autophagic pathways for degradome	[82]
Proteome-wide analysis of chaperone-mediated autophagy targeting motifs.	Cancer, neurodegeneration, and diabetes	Proteome-wide analysis	KFERQ motifs are screened within silico to determine the importance of the acidic and hydrophobic residues in the construction of motifs.	Identification of KFERQ-like motifs	[83]
Autophagy Modulators Profoundly Alter the Astrocyte Cellular Proteome.	Glioblastoma	Proteome analysis	The broader effects of autophagy modulators on host cells are explored.	Mechanisms and effects of Autophagic inhibitors	[84]
Quantitative phosphoproteomic analyses identify STK11IP as a lysosome-specific substrate of mTORC1 that regulates lysosomal acidification.	Fatty liver	Quantitative phosphoproteomic analysis	STK11IP phosphorylation means lysosome can be acidification in face of aberrant autophagy signaling.	Finding novel therapeutic targets	[86]
An integrative multi-omics approach uncovers the regulatory role of CDK7 and CDK4 in autophagy activation induced by silica nanoparticles.	Cancer	Transcriptomic, proteomic, and phosphor-proteomic profiling	Key regulators are identified from the multi-omics data by a new computational prediction algorithm.	The mechanism of cMAK to other nanomaterials	[91]
Multi-omics reveals a relationship between endometrial amino acid metabolism and autophagy in women with recurrent miscarriage.	Recurrent miscarriage	Metabolomic and transcriptomic profiling	Integration multi-omics analysis and RNA-SEQ reveal the signaling pathway of RM.	The relationship between AMPK and amino acid metabolism	[92]
Identification of autophagic target RAB13 with small-molecule inhibitor in low-grade glioma via integrated multi-omics approaches coupled with virtual screening of traditional Chinese medicine databases.	Low-grade glioma	Integrated multi-omics approaches	The autophagy regulatory factors are identified using omics data and the corresponding compounds are identified using virtual screening.	The research and development of new autophagic inhibitors to treat LGG	[93]

Genomics analyses for autophagy

Genomic analysis is the interpretation of phenotype at the DNA level in terms of gene mutation, deletion, amplification, and epigenetic inheritance (DNA methylation). The onset mechanisms of autophagy-related diseases are mostly traced back to abnormalities at the gene level, and identifying the mutated driver genes is key to developing targeted therapies [70]. In silico approach has been applied to chemical genomic analyses and aggresomal clearance assays to reveal the complex molecular mechanisms of autophagy modulators. Principal component analysis (PCA) is a type of multi-source data analysis in which several principal components can be extracted from high-throughput data as representatives to analyze gene expression patterns through reduction and feature selection [71]. PCA and clustering analysis indicated that memantine and clementine induced autophagy via endoplasmic reticulum stress responses. PCA revealed that SMK-17 activated autophagy, which played an inducing role in autophagy via PRKC/TFEB activation, and this was subsequently confirmed by experimental means. Small molecule compounds involving SMK-17 to activate autophagy significantly improve the clearance of abnormal protein aggregates for treating proteinopathies, which are especially relevant in neurodegenerative diseases [72]. Chemical genomic analysis performs classification functions, by which autophagy inducers are arranged into several levels based on the autophagy-inducing signaling pathways. The above-mentioned combination of chemical genomic analysis and PCA approaches shed a new light on discovering autophagy inducers and promoting more efficient targeting of autophagic drug development. For the complexity of mutation-induced pathogenesis, genome-wide mutation characterisation is more reflective of the full picture of genomic alterations in diseased cells than sequencing of targeted autophagy-related genes. For example, systematic integration of whole-genome sequencing technologies, massive pan-cancer population cohorts and data analysis tools not only capture driver mutations but also identify mismatch repair or homologous recombination defects, providing more comprehensive and accurate molecular evidence for research and promoting the clinical application of genomics in targeted interventions or immunotherapy [73]. In the era of big data, coupled with computer sequence analyses, genomic analyses based on massive biological data as good information mining objects can obtain implicit biological tips of knowledge, such as gene function, expression modes, and biological evolution. Mass data processing relies on the development of computer science to develop more analytical algorithms to guide potential therapeutics.

Transcriptomics analyses for autophagy

With characteristics of time, tissue and space specificity transcriptome refers to the collection of all transcriptome products in a cell under a certain physiological condition, including mRNA, ncRNA, rRNA, etc. Numerous studies have shown that the regulation of transcription levels is crucial in the interaction between autophagy and apoptosis, which is also an important aspect of autophagy-related diseases. Circular RNAs (circRNAs), single stranded circular transcripts without 5′ hat construction (m7G) and 3′ Poly (A) tail, promise biomarkers of human diseases because of high evolutionary conservation and abundance [32]. Accumulating evidence has explored the biological mechanisms of circRNA in breast cancer, referring to the context of the correlation between circRNA, cancer callogenesis, and metastasis [30]. The first step was to find the key circRNA and this was achieved through the integration of RNA sequencing, expression difference analysis and pathway enrichment analysis. Findings preliminarily concluded that circROBO1 overexpression inhibits autophagy by inhibiting the transcription of BECN1 to promote cancer progression and liver metastasis of breast cells through feedback loops [74]. Differential analyses based upon RNA-Seq have shown that circNRIP1 can exhibit higher expression in gastric cancer. Combined with its biological function assays, circNRIP1 has been identified as a novel oncogene; thereby providing a new avenue for potential gastric therapeutics [75]. Spatial transcriptome technology bridges the gap between single cell sequencing techniques that make it difficult to measure the relationships between individual cell locations, providing new transcriptome data. Innovative analytical methods to interpret spatial transcriptome data are one of the current frontier issues in the field of bioinformatics. Coupled with machine learning algorithms, spatially resolved transcriptomics (SRT), a novel high throughput sequencing method, has been performed in exploring the spatial heterogeneity of cancer cells to provide insights into dysfunction modes of cells in intact tissues [76] (Fig. 3B). Referring to the study by Jiachen Li et al, the spatial transcriptome data is modelled by creating an adjacency matrix with the distance between cells, inputting graph structure to extract features, obtaining information between local and global features to train a graph convolutional neural network model, predicting the global structure information of nodes, and completing cell clustering based on vectors unique to each cell [77]. The coupling of artificial intelligence and spatial transcriptome technologies provides new ideas for analyzing data. By collating data from model inputs to obtain different feature extractions and build different types of models, this new scheme is expected to be applied to the study of biological functions including autophagy at multiple levels including spatial distribution of gene expression, analysis of cell dynamics and molecular mechanisms of cellular interactions, and to advance clinical trials of therapeutic strategies targeting specific cellular phenotypes and cellular interactions.

Fig. 3.

Omics methods for modulating autophagy. A. RNA sequencing and label-free proteome association to search for differentially expressed autophagy-associated genes, analyze the signaling pathways involved in these genes and reveal the molecular mechanisms by which autophagy dysregulation leads to disease. B. Graphical convolutional neural networks use transcriptomic data to build models to identify cell types, aggregate functionally relevant proteins and analyze key cell subtype interactions.

Proteomics analyses for autophagy

Proteomics detects the abnormal expression of proteins from genetic mutations and copy number abnormalities to identify aberrations of autophagy-related regulators that grasp the crux of targeted therapy resistance for disorders. Proteomic analysis showed that autophagy-related gene 3, a novel protein, targeted P63 to regulate fatty acid metabolism and catalyzed L3 lipidation for abnormal steatosis [78], [79]. Global proteomic analysis of data obtained from the Cancer Cell Line Encyclopedia showed that Parkin, which is involved in protein ubiquitination, has the vital neuroprotection and tumor suppressive roles by regulating mitophagy [80], [81]. an in silico screen of KFERQ-like motifs has been reported to perform through proteome analysis to identify the protein sequences amenable to chaperone-mediated autophagy degradation [82]. Proteome-wide analysis was applied to compare proteomes from several species to elucidate the enrichment results of KFERQ-like motifs. According to the facility, for further analysis, a free web-based resource named KFERQ finder was developed, which can identify KFERQ-like motifs in any proteome [83]. To further elucidate the effects of autophagic regulators on host cells, SOMA (an aptamer-based proteomics platform) has been developed to examine host cell proteins in U-251 astrocytic cells [84]. Quantitative proteomic analysis is used to screen the expression differences of the same proteins acted on via different factors, and the related interference mechanism can be discussed by integrating the functions of screened proteins [85]. Similar to KFERQ finder and SOMA mentioned above, the basic framework of omics analysis requires only fool-proof input of normative data to obtain results, making the visualization of omics data prevalent.

Mechanistic target of rapamycin complex 1 (mTORC1), a serine/threonine kinase that can position on the lysosome surface and plays a central role in regulating cell activities, including mitosis and cellular growth. The activation mechanism of mTORC1 has been revealed, but the downstream signaling output remains unknown. Recently, STK11IP, a substrate of mTORC1 on the lysosome surface, was identified as a new therapeutic target for fatty liver disorders through quantitative phosphorylated-proteomic analyses. Elevated levels of autophagy in STK11IP knockout cells implied that mTORC1 relied on STK11IP phosphorylation to affect autophagy. A combination of biochemical studies, cross-referenced lysosomal proteomic datasets, and mTORC1-regulated phosphoproteomics datasets were analyzed to further elucidate the role of STK11IP in a complex signaling network [86]. Phosphoproteomics analyses have been utilized to identify ULK substrates to reveal the mechanism of VPS34 as an autophagy regulator. PRKAG2 and VPS34 are phosphorylated under nutrient and energy deficiency stimulation to perform biological activities that include stabilizing ULK1 at omegasomes [87]. Using a variety of experimental methods in conjunction with phosphoproteomics, researchers have achieved many breakthroughs including: discovering two new ULK-dependent tables, discovering novel ULK substrates, revealing phosphorylation regulation of VPS15, and the importance of the pseudokinase domains related to recent structural insights [88], [89]. Phosphoromics analysis allows researchers to determine changes in protein phosphorylation levels. If a protein is phosphorylated, then it is most likely to be involved in the function of regulation in the autophagy process. However, how protein phosphorylation affects autophagy is still unknown, therefore, further studies of protein interactions are needed. Compared with other omics methods, phosphorylated proteomic analysis is expensive and the databases that can be linked are limited, resulting in weak systematic analytical capabilities. Although proteomics has made rapid advances in sensitivity and speed, there has been a huge cost behind the scenes. The limitation is the complexity of the technology involved in developing mass spectrometers with high specificity and creating new data analysis methods. Moreover, it is required to operate complex techniques and specialized software skillfully in lengthy workflows. It calls for global openness, cooperation and resource sharing to build accessible and sustainable projects for collecting and sharing various technologies, professional knowledge and experimental data. Strengthening cooperation in interdisciplinary fields is indispensable. Selecting appropriate features from the huge proteomics data to execute model training can obtain novel analysis results related to diseases to further guide clinical.

Multi-omics analyses for autophagy

The multi-omics analysis integrates multiple high-throughput data that can both validate and complement each other at different levels, making it an important technical tool for research in network biology and systems biology. The combined transcriptomic and proteomic analysis allows for a comprehensive measurement of gene expression at the transcription and translation levels respectively, obtaining a panoramic view of the expression and regulation of each step of gene expression and uncovering new results not found by conventional individual omics approach. More recently, new progress has been made in the study of autophagy through multiomics methods. CDK4 and CDK7, two oncogenic kinases, regulate the cell cycle and promote cancer [90]. Combined with computational prediction, silica nanoparticle experiments, and multi-omics approache, new autophagy regulators and the function of CDK7-CDK4 cascade have been found to play a role in activating key transcription factors to induce autophagy [91]. The results of QRT-PCR and RNA-Seq analysis showed that autophagy was overactivated in recurrent miscarriage endometrial samples. The combination of metabonomic and transcriptomic analyses confirmed that abnormal crosstalk between amino acid metabolism and autophagy was a casually related to recurrent abortion [92]. Coupled with the least absolute shrinkage and selection operator algorithms, a novel strategy integrating multi-omics methods and virtual screening has been applied to illuminate how gallic acid negatively acts on RAB13 to inhibit autophagy in low-grade glioma [93]. Genomic methods integrating DNA methylation and gene expression differences were used to analyze the expression differences of autophagy genes in triple-negative breast cancer and enable selection of targets for new therapeutic drugs [94]. Targeted regulation of autophagy, a viable avenue for treating invasive breast carcinoma, highlights SF3B3 and SIRT3 regulating selective autophagy as candidate druggable targets for subsequently healing breast carcinoma patients [94]. In addition to the discovery of new autophagy regulatory factors, multi-omics approaches are also involved in the virtual screening and identification of targeted autophagy regulators. Chemical genomic analysis has a screening function, and when combined with PCA, helps to further reveal the molecular mechanisms of autophagic regulators for accurate treatment of diseases [72]. Summarizing the above studies, transcriptome analysis is common in multi-omics approaches, which are applied to analyze the dynamics of ATG proteins and determine the biological processes in which small-molecule regulators are involved before conducting further experiments (Fig. 3A). Compared with only omics method, multi-omics analyses exploring diverse omics levels of autophagy-related changes provide a systemic clue on elucidating the relationship between autophagy and diseases.

Mathematical models for modulating autophagy

Mathematical modeling, a powerful mathematical method based on numbers and mathematical logic through abstraction and simplification, establishes an approximate description to solve diverse practical problems. Due to the massive datasets available for autophagy-related diseases, research on autophagy by in silico methods has advanced to formulate mathematical models. Autophagy flux describes the measurement of the dynamic characteristics of the autophagy pathway and measurement of such can help in selecting the important variables related to the occurrence of diseases. The experimental methods to accurately measure autophagosome flux are mainly optical microscopes and fluorescent probes, which have some disadvantages, including sources of background signal noise [95], [96]. The transformation of cells from autophagy to apoptosis often results in dyshomeostasis, which can lead to cancer and other diseases. Mathematical models can describe the analytical framework of cell fate determination to assist in predicting and therapeutically interfering with the fate of cells [95], [97], [98]. Mathematical models that portray abstractions of realities have been introduced in the grasping process and applied to patterns of the complex evolution of autophagy. Mathematical models study the ATGs/proteins that control the autophagy process and the underlying associations between autophagy and diseases (Table 4).

Table 4.

Mathematical models for modulating autophagy.

Title	Disease	Method	Feature	Application	Ref
A mathematical model of p62-ubiquitin aggregates in autophagy.	Neurodegenerative diseases	The PDE model	The dynamic process of p62 oligomer level can be described by a model.	The numerical simulations of dynamic stability	[4]
Model-based pathway enrichment analysis applied to the TGF-beta regulation of autophagy in autism.	Autism	Computational modeling, enrichment analysis	System-level model-driven approaches are applied to differentiate between conditions of health and disease in autism.	The monitoring of disease conditions	[63]
Systems-level feedbacks of NRF2 controlling autophagy upon oxidative stress response.	Cancer	Mathematical model, computational modeling	Inducing autophagy through targeting AMPK, NRF2 and mTOR can adapt dynamically changing environments.	The function of systems-level feedback of NRF2	[100]
Prediction of optimal drug schedules for controlling autophagy.	Cancer	Ordinary differential equation model	Locally optimal control strategies from a control engineering perspective are found. Minimize the number of drugs for regulating autophagy.	Monotherapy and multitherapy regiments	[101]
A multimethod computational simulation approach for investigating mitochondrial dynamics and dysfunction in degenerative aging.	Degenerative aging	Agent-based modeling, computational biology	Take an in silico approach to discuss the mitochondrial free radical theory of aging. A multilevel hybrid-modeling paradigm is developed to integrate mitochondrial stress response pathways.	Molecular determinants of degenerative aging	[102]
Computational modeling of the effects of autophagy on amyloid-β peptide levels.	Alzheimer’s disease	Modeling and simulations	The quantitative relationship between amyloid-β peptide (Aβ) levels and autophagy activity is researched. Aβ-regulation failure resulting in elevated levels of autophagy may lead to clogging neurons.	The therapeutic and preventive implications of Alzheimer's disease	[103]

Neurodegenerative diseases are caused by ubiquitinated aggregates whose function remains unclear. In a recent study, a mathematical model based on ordinary differential equations was applied to derive and analyze p62-ubiquitin aggregates in cellular autophagy. To balance the dynamic process of mass change of autophagy markers, the protein p62 and the stability of numerical simulation, three-parameter regimes were determined [4]. In addition, mATG13 dynamics as a part of a dynamic network of autophagy can be captured by mathematical modeling. To respond to starvation and ivermectin, the cells translocate mATG13 to the autophagic puncta to initiate the nucleation of phagophores. Based on the experimental data, combined with mathematical modeling and in vivo imaging studies, the researchers explored the multiple translocations of mATG13 of mitochondrial fragment induction to gain the early membrane rearrangement dynamics of autophagy to provide insight into further regulating it [99]. The linkage between mathematical modeling and computer algorithms extends the utility of data modeling. The regulatory combination of AMPK, NRF2 and mTOR was revealed, and their roles in autophagy induction in response to altering environments was hypothesized via computational modeling [100]. Mathematical models were also introduced in predicting the effects of the molecular targeting drug on cellular fates and autophagy-related disease activities. An ordinary differential equation (ODE) model has been applied to guide a control engineering perspective that extracts insights into optimal drug dosing schedules. An available novel avenue was raised for minimizing the number of drugs to achieve treatment of diseases via sustained targeting suppression of autophagy [101]. Ordinary differential equations were applied to a multilevel hybrid-modeling paradigm to model the dysfunction of aging mitochondrial phenotypes for healing degenerative aging [102]. To study how autophagy affects amyloid-β peptide levels and contributes to Alzheimer's disease (AD), a mathematical model was developed to predict the combined effects of different steps of the autophagy process for Aβ degradation [103]. Differential equation-based models such as delay differential equations (DDE), stochastic differential equations (SDE), and ODE have been used to study the crucial genes/proteins and their underlying dynamics [101], [104], [105]. Mathematical modeling is used to identify targets to compensate for the lack of a therapeutic pipeline targeting binding sites of a protein in the autophagy realm. In mathematical modeling, developing a solid mathematical foundation and modeling intuition may be the important factors to select appropriate parameters which largely determine the quality of the model.

Artificial intelligence (AI) approaches for modulating autophagy

AI means that human intelligence is applied to machines that have the principal computational power to imitate human thinking via efficient algorithms. As a branch of AI, machine learning (ML) based upon deep learning (DL) has been applied in diverse domains, especially in medicine, including disease diagnosis, drug development, and prognosis prediction [106] (Fig. 4). Recently, researching tools and methods based on AI have revealed genetic variants and abnormalities in diseases and have participated in treating several cancers caused by autophagy dysregulation [107]. Drug therapeutics mostly act via protein targets, so finding novel protein targets and corresponding drug development can be beneficial to curing autophagy-related diseases [108]. Studies on autophagy have accumulated diverse and highly accurate big data which is the premise of artificial intelligence to perform feature engineering. With the high-speed progress of AI and further research on autophagy, an abundance of novel tools and methods based on AI have been introduced in therapeutic strategies for autophagy. In this section, we summarize prominent tools based on AI for autophagy research involving autophagy regulatory mechanisms associated with human diseases (Table 5).

Fig. 4.

Artificial intelligence method for modulating autophagy. The process of the deep learning (DL) method includes data acquisition, feature engineering, model building, evaluation, and application. Inputting data related to autophagy can obtain the users’ expected results from the trained model.

Table 5.

Artificial intelligence methods for modulating autophagy.

Title	Disease	Method	Feature	Application	Ref
Targeting the anti-apoptotic Bcl-2 family proteins: machine learning virtual screening and biological evaluation of new small molecules.	leukemia, melanoma, and lung cancer	Machine learning	Two novel prospective inhibitors of the Bcl-2 family are identified by AI technology coupled with experimental methods.	The treatment of cancers by targeting the inhibition of anti-apoptotic proteins	[109]
Machine learning with autophagy-related proteins for discriminating renal cell carcinoma subtypes.	Renal cell carcinoma	Machine learning, quantitative IHC analysis	The basal level of autophagy is applied to tumor discrimination. The K-Nearest Neighbor is applied to quantified numeric data.	The classification of tumors subtypes	[114]
A synergized machine learning plus cross-species wet-lab validation approach identifies neuronal mitophagy inducers inhibiting Alzheimer disease.	Alzheimer disease	Machine learning	The mechanisms by which two lead compounds induce autophagy are deciphered.	Efficient screening of autophagy-induced compounds	[116]
ATTEC: a potential new approach to target proteinpathies.	Huntington disease	Microarray chip, the compound-protein interaction	4 mHTT-lowering compounds are identified by a novel small-molecule screening.	Treatment of diseases by disease-causing proteins	[118]
Dysregulation of apoptosis and autophagy gene expression in peripheral blood mononuclear cells of efficiently treated HIV-infected patients.	Acquired immune deficiency syndrome	Reverse transcription-quantitative PC, machine learning algorithms	Three separate machine-learning algorithms accurately classify peripheral blood mononuclear cell samples.	Treatment of autophagy gene expression disorders	[119]
Amelioration of Alzheimer’s disease pathology by mitophagy inducers identified via machine learning and a cross-species workflow.	Alzheimer’s disease	Machine learning, cross-species workflow	18 small molecules of autophagy-related regulation are identified. Defective mitophagy is related to memory loss in Alzheimer's disease patients.	Therapeutic strategies for autophagy in Parkinson's disease	[137]

AI technologies for autophagic mechanisms

When the expression of Bcl-2 family antiapoptotic proteins in tumor cells is upregulated, identifying specific targets and developing inhibitors are promising therapies for certain tumors. In a recent study, the authors integrated virtual screening based on ML and molecular docking to select two potential lead compounds that were further established by pharmacokinetic analyses [109]. ML was introduced to study the core regulatory machinery that governs the dynamic autophagy process. DeepPhagy, a computer tool for measuring autophagy marker activity, can automatically analyze confocal images and quantitatively sort autophagic phenotypes in Saccharomyces cerevisiae [110]. DeepPhagy, which is expected to be extended to human autophagy phenotype studies, detects autophagy activity automatically and greatly facilitates scientists’ understanding of autophagy dynamic processes under physiological and pathological conditions. Compared with traditional ML strategies, DL algorithms can better extract target features from big data hiding much biological information [111], [112]. To promote AI approach performance, developers can collect manually labeled accuracy datasets and stack fittable layers of convolutional neural networks (CNNs) in a deep learning algorithm. Algorithms determine their core competencies, developing new powerful algorithms to underpin their direct processing of video, rather than just analyzing individual images. In addition, AI is also utilized to study action mechanisms, side effects, and drug repurposing for novel polypharmacological strategies [113].

AI approaches for autophagy-related diseases

The k-nearest neighbor (KNN) algorithm has been introduced in classifying renal cell carcinoma (RCC) subtypes based on the normalized data measured by immunohistochemical images [114]. Their study revealed that the autophagic basal level is a usable parameter for discriminating amongst RCC subtypes. Alzheimer's disease (AD), a common neurodegenerative disease, can be alleviated by inducing autophagy to theoretically degrade disease-causing proteins. Researchers established a virtual screening tool using machine learning and cross-species workflows to find multiple autophagy inducers and combined them with experimental verification to improve the accuracy of AI screening for AD treatment [115], [116]. Huntington's disease, a common progressive neurodegenerative disorder, relates to mutant HTT (mHTT) protein stretch [117]. Autophagosome-tethering compound (ATTEC) is a potential approach to screening based on a small molecule microarray to identify ATTEC compounds that bridge mHTT and LC3, which targets mHTT to phagophores for autophagy-mediated degradation. ATTEC will be applied to develop other small molecule compounds that target activation or inhibition of autophagy for protein disorders and other diseases [118]. Serrano et al. used the software Scikit-learn incorporating ML algorithms to study the effect of dysregulation of autophagic genes for treating HIV-infected patients [119]. This approach enabled the search for differences in the expression of autophagy-related genes between HIV-negative and HIV-positive samples and sought to rank the importance of these genes according to classification criteria. The AI method integrates the training set related to autophagy for extracting input to interpret the autophagy-related gene expression levels in diseased cells, identify disease subtypes, and screen autophagy modulators used to alleviate disorders.

AI methods for autophagic drug discovery

The application of in silico protocols for drug discovery is driven by the need to improve success rates and reduce costs, and the key is target identification, such as targeting autophagy-related gene proteins and autophagy regulatory factors. In the new drug discovery stages, AI predicts drug targets, excluding unsuitable compounds, predicting the relationship between diseases and autophagy activity, and developing new diagnosis and treatment strategies for patients [120], [121]. The ProSelection algorithm yields a superior virtual screening function by filtering out inactive ligands and keeping active ligands[122]. In another study, ML combined with macromolecular docking, particularly useful for malleable proteins, was constructed based on forest classifiers for selecting proper protein structure [123]. According to the principle of similar structure and resembling functional activity, quantitative structure–activity relationship approaches are common in the existing computational models that are used to quickly discover new drugs and assess their chemical risk [124]. However, due to overfitting, activity cliffs, and the inaccuracy of basic experimental data, the prediction results are questionable without experimental validation [125], [126]. Computational modeling is also applied to evaluate autophagy-related compounds for their targets, potential biological activities, and toxicities. Autophagy-related protein inhibitors play an important role in the targeted treatment of autophagy-related diseases. In the case of unclear structures of targets and drugs, computer methods are used to build pharmacophore models based upon the structure-based Glide docking approach to complete molecular docking to discover new inhibitors [127]. Based upon big data available for drug candidates, AI use for drug discovery and drug repurposing also promotes the development of targeted autophagy drugs [128], [129]. Novel coring functions and docking algorithms have been developed to address challenges including protein flexibility, solvation, and prediction of a single target based on multiple structures in computer-aided drug design.

Conclusions and perspectives

Hitherto, researchers have mostly focused on strong correlations between autophagy and diseases, rarely mentioning further relationships among autophagy-related diseases. We think that constituting modules between autophagy and the diseases relevant to autophagy is advantageous to the homologous treatment strategy and flexibility between different modules. In silico approaches can help researchers capture missing pieces and gain more comprehensive insights into developing detailed therapeutic strategies for autophagy-related diseases. The in silico methods are constructed on the basis of the large amount of data provided by the experimental methods. These results obtained by computational approaches need to be confirmed by experimental methods, and those without experimental verification should be taken with caution. As the latest computer method, AI has excellent performance in protein structure prediction, new protein–protein interaction discovery, autophagy dynamic process tracking, drug design, etc. There is no doubt that AI-driven mechanism research, drug discovery, and disease treatment are future innovation hotspots and trends around the world. It is a promising example that the spatial transcriptome technique based on a graph convolutional neural network can be used to identify spatially variable genes, which provides a new technical basis for understanding the relationship between spatial domains and biological functions [130].

Moreover, the other four in silico methods also have unique advantages. The database resource integrates information about autophagy, which facilitates researchers' querying and analysis of data. The systems biology method analyzes the pathogenesis of autophagy disorder from a macroscopic perspective by combining various subnetworks. The omics method excavates the autophagic target around the central law of biology and leads to subsequent effective experiments. Mathematical models formulate mass changes in dynamic processes of autophagy and visualize biological processes. Although in silico autophagy approaches have become available online, facilitating the development of therapeutic strategies for autophagy is hindered by some limitations (Fig. 5).

Fig. 5.

Pros and cons of in silico therapeutic strategies for modulating autophagy. The five methods have their advantages and disadvantages. Databases: Promoting real-time, comprehensive, and accurate data in a database is laborious but worthwhile for providing high-quality data for other methods. Systems biology approaches: Systems biology methods focus on macro-level connections and lack research on local interactions and crosstalk. When more detailed local mechanisms have been explored, the framework of this method is also applied locally. Omics approaches: Combining multi-omics methods and bioinformatics methods, mining targets at the genetic level in diseases has guiding significance for selecting drugs to treat diseases. However, the paradigm of omics is singular, with the expectation of creating richer databases and analyzing omics from more perspectives. Mathematical models: The parameter selection of mathematical modeling determines the quality of the model. Combined with deep learning to select optimal parameters based on big data, it is expected to describe the dynamic process of autophagy more accurately. Artificial intelligence technologies: AI technologies use neural network models to iteratively train high-throughput data, which can greatly reduce the overfitting phenomenon of conventional mathematical models and improve the accuracy of calculation results. In addition, unsupervised learning models can also better capture the subtleties in the dynamic process. The extraction of features in the deep learning process depends on the availability of huge data, the higher the data quality, and the more accurate the modeling.

Compared with the existing biology-related databases, autophagy-related databases contain insufficient information. It is urgent to establish new databases involved in protein–protein interactions, microarrays, microbiology experiments, disease metabolic pathways relevant to autophagy, etc. To integrate autophagy-related factors into a unified conceptual framework, it is essential to develop a platform that provides networks of autophagy-related gene expression profiles, phenotypes, and regulatory factors. The linkage between the database and analysis software will greatly enhance the value of the database by increasing the potential of biological information. Systems biology and network analysis have been performed extensively in studying autophagy, but the stability and percolation of the network need to be improved. The literature has reported the effective application of various novel systems biology approaches combined with computer technologies, but most of them have yet to be applied to the autophagy domain, such as image-based modeling and some analysis packages. The method of omics has been widely used, but functional expansion is limited by formalization. Establishing novel databases is a breakthrough approach to decrease the limitation of omics analysis. With the iteration of experimental approaches and bioinformatics to target autophagy for disease treatment, new tools such as Squidpy, an extensible framework for spatial quantification omics analysis in Python, need to be developed to store, analyze, and interpret omics data [131]. Systematic analysis of various omics requires a large amount of data resources and powerful computer performance to carry out multilayer and complex analysis pipelines. The non-server computing paradigm simplifies the underlying architecture of traditional cloud platforms, enabling more research groups to successfully utilize omics methods [132]. Parameters based on available and high-quality experimental data drive mathematical models that are limited to study parts of the autophagy system. Because of inadequate experimental studies, without high-quality experimental data, new parameter values for modeling remain unknown. The high quality of abundant data is the premise to ensure the accuracy of AI training, and the complexity of autophagy is also a challenge for the application of AI algorithms. Various components of the autophagy pathway have different functioning timespans, and capturing dynamic processes with algorithms is viable in the foreseeable future. The convergence of AI and other disciplines is a source of innovation and studies on the combination of AI and radionics to treat cancer have been reported [133]. Using a part of the functions of Pandomics, a new AI drug target discovery platform was combined with the generative chemistry terrace and the clinical trial prediction platform to advance the discovery of potential druggable targets. This approach further allows for the new drug to enter clinical application research in a very short time. The whole process of the research phases has refreshed the existing records. The unsupervised learning of AI overcomes the disadvantage of manually screening large amounts of data, making it more feasible to screen autophagy inhibitors and activators from compound databases [134]. Prospective validation is a way to predict clinical trial outcomes. Translating research findings into clinical trials with favorable prognoses increases confidence in the paradigm of exploration of biological processes, including autophagy. We expect that AI-based research between autophagy mechanisms and diseases relevant to autophagy will result in paradigm shifts in clinical applications, thereby improving the accuracy of early diagnosis and clinical prognosis prediction for dramatic enhancement of survival rates in patients.

As mentioned above, in silico methods may elucidate how autophagy links to diseases from different angles, but there are overlapping areas of their function, which are mutually validated to improve the accuracy of the results. In silico approaches are used for the co-development of experimental and computational methods for extracting novel insights for autophagy mechanisms, optimizing drug scheduling, and treating related diseases. In addition to performing basic medical research, applications of in silico approaches in translational studies modulating autophagy are rapidly emerging, and the prospect of their integration into therapeutic and clinical trials seems viable.

Compliance with Ethics Requirement

This review does not involve any studies with human or animal subjects.

Credit author statement

BL and HYY designed and directed the review. LFW And WKJ contributed to collecting the resources, designing the figures, and writing the manuscript. All authors critically read and approved the final manuscript.

Funding

This work was supported by grants from National Natural Science Foundation of China (Grant No. 82172649), and National Key Research and Development Program of China (Grant No. 2021YFE0203100).

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Biographies

Lifeng Wu received her B.S. degree in Bioscience from Guizhou University in 2021. Now, she is a postgraduate student at State Key Laboratory of Biotherapy and Cancer Center of West China Hospital in Sichuan University. Her research interests include in silico autophagic target identification and drug development.

Wenke Jin received her bachelor's degree from the Ocean University of China in 2019. Now, she is a Ph.D. student at State Key Laboratory of Biotherapy and Cancer Center of West China Hospital in Sichuan University. Her research interests include autophagic targeted drug discovery and relevant antitumor mechanisms clarification via integrating multiple bioinformatics methods.

Haiyang Yu received his Ph.D. degree from Dong-A University, in 2012. He is a Professor of Tianjin University of Traditional Chinese Medicine, since 2015. Up to present, he has published more than 40 papers in journals such as Nature Communications, Journal of Hematology & Oncology, etc. His research is interests are in exploring new druggable targets and new autophagic mechanisms of anti-tumor drugs based upon small molecules from traditional Chinese medicine and natural products.

Bo Liu received his Ph.D. degree in Bioinformatics (drug design) from School of Life Sciences, Sichuan University in 2010. In 2012, he joined the faculty of State Key Lab of Biotherapy of West China Hospital in Sichuan University as a full professor. His research interests include in silico autophagic target identification and drug discovery.