Comparative pharmacoinformatic and quantum descriptor insights from BFM/GBTLI guidelines to phase I/II compounds for acute lymphoblastic leukemia (ALL)

Abstract

Acute lymphoblastic leukemia (ALL) remains the most common pediatric malignancy worldwide. Standard protocols such as BFM and GBTLI rely on long-established cytotoxic agents, yet novel targeted compounds have recently entered phase I/II trials. Despite these advances, no prior study has systematically compared the pharmacokinetic, ADMET, and quantum descriptor profiles of protocol-based drugs versus emerging clinical-phase agents. This study addresses that gap by integrating pharmacoinformatic and quantum-chemical approaches to highlight differences with potential clinical implications. We retrieved all small-molecule drugs from the BFM/GBTLI 2009 protocols and a representative set of phase I/II investigational compounds for pediatric ALL. In silico tools were used to assess physicochemical properties, ADMET (absorption, distribution, metabolism, excretion, and toxicity) profiles, and quantum chemical descriptors. We evaluated physicochemical and pharmacokinetic properties, including solubility, permeability, metabolic liabilities, and toxicity risks. Quantum chemical descriptors were calculated with density functional theory (DFT) to assess molecular reactivity (HOMO, LUMO, gap, dipole moment, electrophilicity). Multivariate analyses were applied to compare and cluster drug profiles. The comparative analysis revealed significant variability between guideline and clinical-phase compounds. Clinical-phase compounds generally exhibited higher molecular weight and lipophilicity, together with greater variability in permeability and solubility-related descriptors, indicating potential formulation and bioavailability challenges. Several investigational agents were identified as P-gp substrates and hERG inhibitors, suggesting increased risk of efflux-mediated resistance and cardiotoxicity. Quantum chemical analysis revealed that phase I/II compounds (e.g., Pelabresib, Molibresib) displayed smaller HOMO–LUMO gaps and higher electrophilicity, consistent with higher theoretical reactivity, whereas guideline drugs (e.g., Vincristine, Methotrexate) showed more stable electronic profiles. Cluster analysis confirmed distinct grouping between guideline and clinical-phase compounds. This in silico comparison integrates pharmacoinformatic and quantum descriptor analyses of established and emerging ALL therapeutics. By revealing key differences in drug-likeness, ADMET, and electronic reactivity, the study provides a comparative framework that may support the prioritization, optimization, and clinical translation of next-generation therapies for pediatric ALL.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-36374-9.

Introduction

Pediatric cancer is a major global health challenge. An estimated 400,000 new cases per year result in more than 105,000 deaths worldwide ^1,2. Although childhood cancer accounts for only ~ 1% of all malignancies, they are the leading cause of disease-related deaths in individuals aged 1–19, with leukemia being the most common form ³. Survival rates for childhood cancer vary greatly from country to country. Survival exceeds 80–90%% in high-income regions, but remain bellow 30% in low- and middle-income countries due to delayed diagnosis and limited access to effective care. Initiatives such as the World Health Organization’s Global Initiative for Childhood Cancer aim to increase the global childhood cancer survival rate to at least 60% by 2030 ¹. Common pediatric cancers include leukemia, particularly acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML), followed by brain tumors, neuroblastoma, Wilms tumor, lymphomas, sarcomas, and retinoblastoma ⁴.

ALL is the most frequent cancer in children under 15 and accounts for ~ 75% of leukemia cases in this age group. Survival rates for juvenile acute lymphoblastic leukemia (ALL) have increased dramatically thanks to breakthrough treatments and supportive care. These rates currently surpass 90% ^5,6. Despite these improvements, there are still challenges such as life-threatening side effects of chemotherapy and poor survival rates in relapsed or refractory ALL cases ⁷. The standard treatments, including chemotherapy, surgery, and radiotherapy, are essential in ALL care. Additionally, palliative care plays a vital role in enhancing the quality of life for both patients and their families ^7,8. However, new chemotherapeutic agents face substancial obstacles, including systemic toxicities and drug resistance. These issues directly impact the effectiveness of therapy and the overall well-being of the patients ⁸.

Early evaluation of chemical toxicity and pharmacokinetic liabilities is therefore crucial. The integration of computational methods at an early stage of drug discovery is essential for predicting ADMET properties (absorption, distribution, metabolism, excretion, and toxicity), which largely determine the pharmacokinetic profile of a drug ⁹. Additionally, density functional theory (DFT) calculations and molecular modeling provide significant insights into the electronic structure and potential reactivity of molecules, aiding in the early identification of compounds of interest ^10,11. These computational tools enable the efficient investigation of large chemical libraries, significantly reducing the time and cost of experimental ADMET research ¹². Optimizing lead compounds with attractive pharmacokinetic and safety properties expedites drug discovery and enhances the probability of success in clinical trials ¹³.

Recent studies underline the need for comprehensive investigations into drug compliance, interactions, bioavailability and sensitivity to optimize personalized treatment strategies. This includes exploring and comparing different treatment protocols, such as the Berlin–Frankfurt–Münster (BFM) protocol, to improve outcomes and reduce complications in childhood ALL ¹⁴. These protocols emphasize the importance of differentiated approaches based on prognostic factors and the use of supportive measures to manage toxicity, including myelosuppression, infections, hepatotoxicity, and skin/mucosal toxicity ^15–18.

Within this framework, and to ensure clinical relevance in subsequent in silico analyses, the chemotherapeutic agents were selected from the ALL-IC BFM 2009 and Brazilian Group for Childhood Leukemia Treatment (GBTLI 2009) protocols, both recognized as international and national standards of care for pediatric ALL, achieving 5-year survival rates exceeding 90% in high-income settings. The drug lists of these protocols represent the cornerstone of ALL therapy and thus serve as a clinically relevant baseline for comparison. Conversely, we included a set of phase I/II investigational compounds that are currently being explored in ALL clinical trials. We selected pediatric ALL as our clinical setting because the ALL-IC BFM and GBTLI protocols constitute well-established standards of care, offering a clearly defined set of guideline drugs, while several targeted small molecules are currently in phase I/II trials for this population. This combination provides a rich and socially impactful context in which to benchmark emerging agents against protocol-based therapies.

However, despite these established regimens and emerging agents, a major translational gap remains. This study addresses this gap through a comparative and predictive in silico analysis integrating ADMET profiling with quantum chemical descriptors. Rather than developing new prediction algorithms, our contribution is to integrate established pharmacoinformatic servers with DFT descriptors into a unified comparative framework. This framework contrasts the drug-likeness, ADMET profiles, and electronic reactivity, providing a comparative resolution not previously explored in this context.

Together, this integrated comparison aims to advance pharmacoinformatic knowledge by revealing actionable physicochemical and ADMET discrepancies between standard-of-care agents and emerging clinical-phase compounds, thereby providing a translational scaffold for candidate prioritization, formulation optimization, and early identification of pharmacokinetic or toxicity liabilities in pediatric ALL.

Methods

Data collection for chemotherapy compounds

For this comparative study, we retrieved molecular compounds listed in the ALL-IC BFM 2009 and the GBTLI 2009 pediatric ALL guidelines, as well as from the Hyperfractionated Cyclophosphamide, Vincristine, Doxorubicin, Dexamethasone (HYPER-CVAD) regimen ¹⁹. Cytarabine and daunorubicin, core agents of the “3 + 7 Scheme” widely used in the AML therapy, were also included. Clinical phase compounds were identified through a comprehensive literature review ^20–24, followed by data collection from established chemical and pharmacological databases. Structural and pharmacokinetic information was obtained from PubChem (https://pubchem.ncbi.nlm.nih.gov/) ²⁵, DrugBank (https://go.drugbank.com/) ²⁶, and ChemSpider (http://www.chemspider.com/) ²⁷ (Table 1). Phase I/II investigational compounds were retrieved through systematic searches of clinical trial registries and peer-reviewed literature up to the end of the study period. Inclusion criteria required the evaluation in hematologic or lymphoid malignancies, fully defined molecular structures available in public repositories and suitability for ADMET and DFT-based descriptor computation. This process yielded 16 clinical-phase molecules and 10 guideline agents.

Table 1.

Describes the inclusion and exclusion criteria of the retrieved compounds.

Criteria
Inclusion	Exclusion
Guideline compounds	Protein compounds
Compounds reviewed in the literature in clinical phase I and II	Compounds in clinical phase III onwards
Compounds for the treatment of onchometological conditions	Compounds not related to the treatment of neoplastic conditions
Compounds present in reliable and validated databases	Compounds with incomplete data

Drug-likeness rules

The physicochemical properties of the compounds were assessed in accordance with 9 drug-likeness rules/filters developed by pharmaceutical companies, namely Golden Triangle and Egan’s rule, Lipinski’s Rule of Five (RO5) and Veber’s rule, GSKA 4/400 and Pfizer 3/75, Drug-Like soft, and Oral Property Space ^28,29. To conduct this assessment, we utilized the fafdrugs4 (https://fafdrugs4.rpbs.univ-paris-diderot.fr/filters.html) ³⁰, SwissADME (http://www.swissadme.ch/) ³¹ and Molinspiration (http://www.molinspiration.com/) ³² servers. These tools employ a combination of calculations based on physicochemical properties and the identification of toxicophores (potentially toxic chemical groups) to pinpoint compounds with ADMET profiles that are closer to the desired ideal, thereby selecting high-quality candidates.

Fafdrugs4 has consistently handled a significant workload, with more than 15,000 registrations in 2016 and an annual range typically spanning from 10,000 to 30,000 ³⁰. SwissADME is widely utilized as a web tool for rapid prediction of physicochemical properties ³¹. Meanwhile, Molinspiration, known for its extensive citation record of over 4500 in scholarly journals and processing more than 80,000 molecules each month, has consistently delivered reliable results ³². It is essential that the parameters associated with these rules lie within the optimal range, which encompasses the maximum and minimum extreme zones. The descriptor ranges for each rule are provided in the Table 2.

Table 2.

Drug-likeness rules applied.

Rules of medicinal chemistry applied
Golden triangle and Egan’s rules	Used to estimate the oral bioavailability andassimilation of drug candidates. This rule implies that the closer a molecule is to the triangle’s center, the greater its oral absorption and clearance (Golden triangle rule: MW1 200 to 450, LogD2 − 2 to 5. Egan’s rules: LogP3 − 1 to 3, tPSA4 ≤ 132)
Lipinski’s and Veber’s rules	Lipinski’s rule evaluates the probability of a chemical compound exhibiting chemical and physical properties that would render it a potential orally active drug in humans. Simultaneously, Veber’s rule examines the parameters associated with favorable oral bioavailability in a drug (Lipinski’s rule: MW ≤ 500, LogP 0 to 5, HBA4 ≤ 10, HBD5 ≤ 5. Veber’s rule: tPSA ≤ 140, number of rotatable bonds ≤ 10, and HBA + HBD6 ≤ 12)
Drug-Like Soft	The Drug-Like Soft filter assesses the descriptor values of physicochemical properties derived from 916 FDA-approved oral drugs, which were computed using fafdrugs4 and obtained from the e-Drugs3D library (MW 100–600, LogP − 3 to 6, HBA ≤ 12, HBD ≤ 7, tPSA ≤ 180, Rotatable Bonds ≤ 11, Rigid Bonds ≤ 30, Rings ≤ 6 Max Size System Ring ≤ 18, Carbons 3 – 35, Straight Atoms 1 – 15, H/C Ratio 0.1 to 1.1, Charges ≤ 4, Total Charges − 4 to 4)
GSK 4/400 and Pfizer 3/75	The rule aims to predict toxicity and ADMET7 profile by examining the physicochemical properties of a candidate drug molecule. GSK rule refers to the compounds that have a more favourable ADMET profile (LogP < 4, MW < 400), while Pfizer rule identifies the compounds that are approximately 2.5 times more likely to be toxic than to be clean (LogP > 3, TPSA < 75) 33
Oral Property space	An extensively employed tool at Bayer is the Oral PhysChem Score, which is derived by conducting a Principal Component Analysis (PCA)8 on the 15 primary physicochemical descriptors of the user’s compound. This score is then compared to two distinct oral sub-libraries obtained from eDrugs and DrugBank 34

¹ MW: Molecular Weight; ²LogD: Distribution Coefficient; ³LogP: Partition Coefficient; ⁴tPSA: Topological Polar Surface Area; ⁵HBA: Hydrogen Bond Acceptors; ⁶HBD: Hydrogen Bond Donors; ⁷ADMET: Absorption, Distribution, Metabolism, Excretion, and Toxicity; ⁸PCA: Principal Component Analysis.

ADMET properties

To validate the prediction of drug-likeness results and estimate pharmacokinetic features (Table S2), multiple servers were utilized, including ADMETlab 2.0 (https://admetmesh.scbdd.com/) ³⁵, Admetsar (http://lmmd.ecust.edu.cn/admetsar2) ³⁶, SwissADME (http://www.swissadme.ch/) ³¹, pkcsm (http://biosig.unimelb.edu.au/pkcsm/prediction) ³⁷, and Pred-hERG (http://predherg.labmol.com.br) ³⁸. All these tools, including the ones listed in Section “Drug-likeness rules”., are commonly reported on the literature for ADMET analysis due to their reportedly extensive data sets available on their servers, allowing for a wide array of drug properties comparisons and reliable results.

The Pred-hERG tool generates a representation of the molecule, where green areas indicate a high probability of hydrogen bonding between the molecule and the hERG channel, while pink areas indicate a low probability of interaction (Figs. 1 and 2). ADMETlab 2.0 uses advanced mathematical algorithms to evaluate ADMET profiles of chemicals using a comprehensive database of approximately 288,967 compounds and 31 optimized QSAR models ³⁵. These algorithms contribute significantly to optimizing efficiency and accelerating important tasks in chemistry and drug development ³⁵. Similarly, the Admetsar server includes an extensive collection of over 210,000 experimental data points for 96,000 compounds and 27 computer models ³⁶.

Fig. 1.

Overview of ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) parameters assessed in this study, including key transporter-related endpoints such as P-gp substrate/inhibitor status, permeability metrics, metabolic interactions, and toxicity predictors.

Fig. 2.

Radar charts from ADMETlab2 illustrating the physicochemical properties of molecules in clinical phases I/II. The blue line represents the molecular values, while the yellow/red areas indicate the appropriate/inappropriate limits. Molecular structures are shown below each radar chart, providing a visual representation of the chemical compounds analyzed.

The pkcsm server has two main sets of descriptors for predicting the physicochemical properties of compounds: graphical signatures at a distance and general properties of ligands ³⁷. The server uses machine learning methods. The scanning algorithm analyzes 30 predictors divided into main classes: absorption, distribution, metabolism, excretion, and toxicity. Except for the Pred-hERG server, which is specifically used for the early detection of potential human Ether-à-go-go-Related Gene (hERG) blockers and non-blockers in a curated library of 5,984 compounds, these servers are dedicated to predicting ADMET endpoints. The assessed parameters encompass models that are based on drug-like molecules, compounds, and experimental measurements, providing ADMET endpoint predictions.

Transporter-related endpoints were restricted to those consistently implemented across servers, with a particular focus on P-glycoprotein (P-gp) substrate and inhibitor predictions, given their central role in drug efflux and multidrug resistance in ALL. Solubility-related descriptors were computed using the default settings of each server, which estimate intrinsic or aqueous solubility under near-physiological conditions (approximately pH 7.4). These settings were applied consistently to all compounds.

Finally, statistical analyzes were performed using R 4.3.0 and Rstudio software version 2023.06.1 to represent the physical–chemical, pharmacological and toxicological characteristics of the compounds in the two groups in the form of a clustered heatmap (Fig. 3). The “circlize” and “complexheatmap” libraries were used for this, with categorization as a pre-processing of data for better visualization and comparative analysis. Furthermore, the properties were ordered by the power of differentiation of the 2 groups through a machine learning approach by Random Forest (Fig. 4) through the “randomForest” and “caret” libraries. All properties are archived in Supplementary Tables S1 and S2, which summarize physicochemical and medicinal chemistry descriptors, as well as ADMET parameters (Fig. 5). It is important to note that, in this study, predicted properties were used comparatively between compound groups within a single, internally consistent in silico framework. Systematic experimental validation of each predicted descriptor was beyond the scope of this work and is identified as an important direction for future studies.

Fig. 3.

Radar charts from ADMETlab2 illustrating the physicochemical properties of molecules in the guidelines. The blue line represents the molecular values, while the yellow/red areas indicate the appropriate/inappropriate limits. Molecular structures are shown below each radar chart, providing a visual representation of the chemical compounds analyzed.

Fig. 4.

This heatmap illustrates the clustering of various compounds based on their activity profiles across different biological assays. Each row represents a different assay, while each column corresponds to a specific.

Fig. 5.

Performance and Feature Importance Analysis of the Random Forest Model. The left graph illustrates the error rate of the Random Forest model as the number of trees increases, demonstrating the model’s learning progression and stability. The right graph depicts the feature importance ranking, highlighting the most significant variables influencing the model’s predictions, such as P-gp I inhibitor, hERG I inhibition, and Half Time (t1/2).

Calculations of the chemotherapy compounds

Optimization of the geometry

All computational analyzes were performed using BIOVIA Discovery Studio 2023 v.22 (Windows version) installed on a dedicated server of the Universidade Federal do Rio Grande do Norte (UFRN). This server was equipped with 12 processors (model and specifications), 64 GB RAM and a 2 TB SSD for efficient data processing and computation. Discovery Studio v.22 was installed on a dedicated workstation within the UFRN network to ensure access to all computing resources. The installation process followed the guidelines provided by BIOVIA, with particular care taken to remove previous versions to avoid conflicts. Additionally, Pipeline Pilot v.2023 was installed as a prerequisite for DS and enables the automation of computational workflows. This setup was crucial for accurately replicating the study’s computational environment.

Ligand structures were systematically retrieved from the specified databases in .sdf format. Initial ligand preparation was performed using the “Clean Geometry” tool in DS, which was set to the default parameters. This step ensured that all structural anomalies were removed. The geometry optimization was performed with the “Minimize Ligands” module, using CHARMm force field parameters with default settings. The aim of this optimization was to achieve a stable conformation with low energy suitable for further analysis. The detailed steps included the execution of “Quick Minimization”, followed by “Full Minimization” for each ligand. Additionally, the “Smart Minimizer” algorithm was applied within the minimization process to ensure optimal geometric configurations. This preparation protocol ensured that all ligands were standardized, maintaining consistency in subsequent analyzes.

A detailed conformational search was performed using the “Search Small Molecule Conformations” tool. The settings were set to “Add Rotable Bonds” and the “Generate Conformations” module in DS was used using the CHARMm force field. A maximum of 500 conformers per molecule were considered, with an RMSD cutoff of 0.5 Å to eliminate redundant conformers, with the method set to “BEST”. The energy threshold was set to 3 kcal/mol to identify low energy conformations. The conformer with the lowest energy from the generated pool was selected for subsequent quantum chemical calculations to ensure that the most stable molecular geometry was analyzed.

The optimization calculations were done with Gaussian 16 software and density functional theory (DFT) at the B3LYP/6-31G(d,p) basis set. This was chosen because it is a good balance between how much it costs to run and how well it predicts electronic structures. Additionally, the PBE functional was employed for its accuracy and computational efficiency, with grid settings set to “FINE” for a detailed investigation of electronic properties. The solvent effects were modeled using a water-solvent model that reflects the biological conditions under which the molecules exert their effects. The “DMol3 properties” option was set to “ALL” to ensure a comprehensive collection of quantum chemical descriptors.

Quantum chemical descriptors

DFT is a widely employed computational quantum mechanical modeling technique, extensively applied in molecular optimization. It provides detailed insights into the behavior of electrons within molecules and their impact on key quantum mechanical parameters and descriptors, such as HOMO, LUMO, GAP, electron affinity (A), molecular hardness (η), softness (σ), chemical potential (μ), electronegativity (χ) and electrophilicity index (ω) (Table S6), as well as electrostatic potential maps and molecular orbital diagrams that facilitate the analysis of potentially reactive sites and electronic distribution (Table S5 and Figs. 6, 7, and 8). Geometry optimization was performed using PBE for robustness and efficiency, whereas HOMO/LUMO energies and global descriptors were computed at the B3LYP level due to its established accuracy in orbital energetics. This complementary approach ensured structural reliability while enabling accurate interpretation of reactivity parameters. These descriptors provide insights into the ligand reactivity and stability, which are crucial for understanding their interaction potential and corroborate findings reported in other studies ^39,40. The mathematical formulations for these parameters are delineated as follows ⁴¹:

1

2

3

4

5

6

7

8

Fig. 6.

Maps of the electrostatic potential of each molecule together with quantitative information on its total energy (Ha), binding energy (Ha), dipole mag, dielectric sol energy (Ha), solvation energy (Ha), surface area and cavity volume.

Fig. 7.

HOMO–LUMO diagram of frontier molecular orbital.

Fig. 8.

HOMO–LUMO diagram of frontier molecular orbital × 2.

Results and discussion

The comparative nature of this work provides an interpretative scaffold for rational design. Mapping phase I/II agents onto the physicochemical and ADMET landscape of guideline drugs highlights deviations in solubility, permeability, transporter engagement, metabolic burden, and electronic reactivity. These differences can guide optimization—through medicinal chemistry or formulation strategies—to bring new candidates closer to the pharmacokinetic and safety profiles that underpin current standard-of-care regimens.

Compounds were screened using stringent inclusion and exclusion criteria to ensure the selection of relevant and high-quality candidates (Table 1). Included compounds were required to appear in validated databases, be reported in phases I and II clinical literature, and be intended for onco-hematological indications. Compounds wew excluded if they were protein-based, in clinical phase III or later, unrelated to cancer treatment, or had incomplete data.

This rigorous filtering process resulted in a selection of 26 compounds, of which 16 belong to groups of clinical phases I or II trials, and 10 are guideline compounds. For this descriptive and comparative study, we retrieved molecular compounds from the BFM 2009 guidelines (ALL-IC, BFM 2009), the GBTLI 2009 guidelines (GBTLI—Brazilian Society of Pediatric Oncology, 2009), and the HYPER-CVAD regimen. Notably, this included key molecules such as cytarabine and daunorubicin, primary constituents of the “3 + 7 Scheme” widely used in AML treatment. Subsequently, we conducted a comparative analysis between the compounds from clinical phases I or II and those categorized as guideline compounds, providing a thorough assessment of their characteristics and potential therapeutic applications. These guidelines establish important standards in the treatment of onco-hematological disease and can serve as valuable references when selecting and developing potential drug candidates (Table S0).This set reflects all eligible small-molecule agents that met our predefined inclusion criteria within the time frame of the literature search. Future applications of this framework may include larger panels of launched targeted therapies or additional biological contexts to further explore these patterns.

Drug-likeness and pharmacokinetics analysis

We conducted an analysis to assess the drug-likeness of molecular chemotherapeutic compounds currently in phase I or II clinical trials for treating various onco-hematological conditions (Fig. 1). We compared these compounds to drugs recommended in the BFM 2009 and GBTLI 2009 guidelines, as well as to established treatment regimens like HYPER-CVAD and the 3 + 7 scheme (Figs. 2 and 3). Our analysis focused on evaluating their structural and pharmacological properties relative to extensive molecular databases.

We observed structural diversity among drugs from clinical trials (Fig. 1) and guidelines for the treatment of onco-hematological conditions (Fig. 2), as reflected, for instance, in the size of the ring systems and the molecular weight (MW) of the compounds. However, despite the significant variability among the analyzed molecules, all compounds met the criteria of drug-likeness rules for these parameters (Max Size System Ring ≤ 18 and a molecular weight ≤ 500 kDa). It is noteworthy to mention that ring systems play a highly influential role in determining the shape, electrostatics, and often bioactivity of compounds ⁴². Nevertheless, there is no universally recommended specific value for this parameter because it depends on the desired properties of the target molecule and the balance between various pharmacokinetic and pharmacodynamic factors. In general, a molecule with a smaller ring system is often preferred as it can exhibit better solubility, bioavailability, and may be easier to optimize for safety and efficacy ^42,43. In some other situations, molecules with larger ring systems may be necessary or desired, depending on the specific mechanism of action and therapeutic target. For example, certain complex molecules, such as some antibiotics, may have larger ring systems due to their intricate activity ⁴⁴.

The group of drugs in the guidelines or in clinical trials share some common physicochemical characteristics, such as neutral formal charge, high molecular flexibility and the ability to form hydrogen bonds (Figs. 1 and 2). These characteristics are essential for the compounds to interact effectively with biological targets in onco-hematological diseases, such as leukemia and lymphoma ^45–47 Nevertheless, we noted significant disparities in the distribution coefficient (LogD) and the octanol–water partition coefficient (LogP). Notably, the compounds undergoing clinical trials exhibited elevated LogD and LogP values (Fig. 1) when compared to the majority of compounds recommended in leading guidelines for the treatment of onco-hematological conditions (Fig. 2). In fact, only the recommended drugs dexamethasone and vincristine demonstrated LogD and LogP values like those of the compounds in clinical trials. It is well-established that during the development of new drugs, it is crucial for candidate compounds to demonstrate elevated LogD and LogP values ^45–47. These metrics serve as significant indicators of lipophilicity, a pivotal physicochemical property that directly impacts ADMET, solubility, permeability, and drug affinity ^48,49.

Across both groups, Caco-2 and MDCK permeability values displayed substantial variability. Classic agents such as 6-mercaptopurine and cyclophosphamide showed relatively high Caco-2 permeability, consistent with their established oral use, whereas several phase I/II compounds (e.g., Molibresib, RO6870810) combined high lipophilicity with predicted high permeability. This pattern suggests that, for many of the newer agents, oral bioavailability is likely to be limited by solubility rather than membrane permeation per se, reinforcing the need for enabling formulations rather than purely structural changes in some cases.

Our methodology of in silico analysis of the ADMET profile has proven to be very important for drug development and optimization, as shown in this work ⁵⁰, which demonstrates the efficacy and safety of modafinil and caffeine in improving alertness and cognitive performance during periods of limited sleep deprivation, thus contributing to the strategic selection of stimulants to increase alertness in critical areas such as aviation. In addition, despite the urgency and initial evidence of efficacy and safety, the approval of tecovirimat for the treatment of monkeypox required further investigation to assess its safety profile. The in silico study of the ADMET profile highlighted concerns regarding potential liver, respiratory and renal damage and carcinogenic risks associated with tecovirimat ⁵¹. These results demonstrate the need and usefulness of this type of assessment, particularly for drugs in emergency health situations.

ADMET properties were predicted through drug-likeness analysis and the data were presented as a grouped heatmap (Fig. 3), in order to facilitate the visualization of how these properties are distributed among drugs from clinical trials or guidelines for the treatment of onco-hematological diseases. This heatmap allows to visualize which properties do not completely distinguish the analyzed compounds, indicating the present and absent properties in each one, as well as those that tend to vary between the compounds. Then, the properties were classified according to the power of distinction of the two groups by the random forest method (Fig. 4). In this way, we were able to characterize several compounds through the “drug-likeness” approach, considering the most prominent properties among them, with reference to the impact analysis in the comparison between molecules in phase I and II clinical studies and those present in well-defined guidelines ⁵².

The comparison of ADMET characteristics between pediatric cancer chemotherapeutics in guidelines and those in clinical phases (I and II) reveals several critical discrepancies worth discussing. Chemotherapeutics in clinical phases exhibit lower water solubility (more negative Log mol/L values) compared to those in guidelines. Notably, drugs like Molibresib (-4.785) and RO6870810 (-4.32) show significantly lower solubility than their guideline counterparts (Table S1). Lower solubility may impact drug bioavailability, necessitating formulation strategies to enhance solubility, ensuring effective delivery and absorption in pediatric patients ⁵³. Consequently, these newer, often more lipophilic agents may require advanced formulation strategies, such as, but not limited to, nano-suspensions or amorphous solid dispersions, to achieve adequate systemic exposure and distribution. This computational insight could serve as an early flag for significant downstream development challenges that must be overcome for clinical success. There is considerable variability in Caco2 permeability among both groups. The 6-Mercaptopurine (1.202) and Cyclophosphamide (1.541) have high permeability in guidelines, whereas drugs like Molibresib (1.069) and RO6870810 (1.285) also show high permeability in clinical phases. High permeability suggests good potential for oral absorption, which is crucial for pediatric patients to avoid invasive administration routes. High permeability is mandatory for effective oral bioavailability, as it allows the drug to efficiently cross the intestinal barrier and reach the bloodstream, potentially reducing the reliance on more invasive intravenous administration in pediatric patients ⁵⁴. Drugs with low permeability like Citarabine (-0.131) may require alternative administration strategies or permeability enhancers (Table S1).

Some drugs in clinical phases, such as Mivebresib and ABBV-744, achieve 100% absorption, surpassing most guideline drugs. Conversely, drugs like Methotrexate (16.001%) and Leucovorin (2.576%) show poor absorption in guidelines. High absorption rates in clinical phase drugs indicate efficient intestinal uptake. Oral bioavailability shows significant variation. For instance, CYC065 (0.94) and Flavopiridol (0.975) exhibit high F(20%) values in clinical phases, while guideline drugs like Vincristine (0.996) and 6-Mercaptopurine (0.973) also show high bioavailability. Disparities in Madin-Darby canine kidney cells (MDCK) permeability are evident, with drugs in clinical phases like Pracinostat (4.72E-05) showing much higher permeability than many guideline drugs. High MDCK permeability is indicative of good cellular uptake, which is crucial for ensuring that pediatric cancer chemotherapeutics reach therapeutic concentrations within target cells ⁵⁵.

The ABBV-744 and Molibresib drugs, are both P-gp (P-glycoprotein) substrates and inhibitors, unlike some guideline drugs. Being a P-gp substrate can lead to drug efflux and reduced effectiveness, whereas inhibitors can block this effect and enhance drug retention within cancer cells. The dual role in clinical phase drugs suggests potential for overcoming P-gp mediated drug resistance, which is a significant concern in pediatric oncology. There is minor variation in skin permeability, but drugs like Dexamethasone (-4.004) and Vincristine (-3.914) from guidelines show significantly lower permeability compared to most clinical phase drugs. The Absorption of chemotherapeutic drugs in pediatric ALL patients is crucial for treatment efficacy ^2,22,54. Pediatric cancer treatment, including ALL, faces challenges due to interindividual variability in drug response, partly attributed to drug transporters affecting drug disposition ^3,6,56. Understanding factors like drug-metabolizing enzymes and drug transporters is essential for optimizing drug exposure and minimizing toxicity in pediatric patients ⁵⁷. Chemotherapeutic agents used in childhood cancer therapy, such as antimetabolites and alkylating agents, interfere with DNA synthesis and replication, highlighting the importance of drug absorption and cellular uptake in achieving therapeutic outcomes ⁵⁷. Compounds predicted as P-gp substrates or inhibitors may influence multidrug resistance patterns in ALL, as P-gp overexpression contributes to cytotoxic drug efflux. Likewise, compounds with strong predicted CYP3A4 interactions may undergo variable metabolism in children, whose CYP3A4 maturation and ontogeny differ from adults, with implications for exposure and toxicity. hERG inhibition predictions further highlight potential cardiotoxicity, a known concern in pediatric regimens.

The intensity of the color represents the level of activity, with red indicating higher activity and blue indicating lower activity. The top dendrogram shows the hierarchical clustering of compounds, while the left dendrogram shows the hierarchical clustering of assays. Compound groups and assay groups are distinguished by the colored bars on the top and left sides of the heatmap, respectively.

Regarding Distribution, Fraction unbound (Fu) and Protein Plasma Binding (PPB) exhibit the most substantial discrepancies. For example, ABBV-744 shows a high Fu of 91.9%, indicating a higher availability of the active drug, which can enhance therapeutic efficacy but also increase the risk of toxicity. In contrast, Daunorubicin has a lower Fu of 15.9%, suggesting extensive plasma protein binding and potentially reduced immediate efficacy. Similarly, PPB values range widely, with compounds like ABBV-744 and 6-Mercaptopurine showing nearly complete binding at 99.33% and 98.87%, respectively, while others like Daunorubicin demonstrate much lower binding at 94.43%. These differences are crucial as high PPB can prolong drug action but might necessitate careful monitoring to avoid adverse effects. The Volume of Distribution (VD) also varies, impacting drug distribution and therapeutic reach. ABBV-744, with a VD of 2.576 Log L/kg, indicates extensive tissue distribution, potentially enhancing therapeutic outcomes by reaching more affected areas. Conversely, Daunorubicin, with a VD of 1.01 Log L/kg, suggests more confined distribution, which could limit systemic exposure and reduce side effects but may also restrict its efficacy in targeting widespread leukemic cells. The Blood Brain Barrier (BBB) and Central Nervous System (CNS) permeability for most compounds indicate minimized central nervous system side effects, crucial for pediatric patients.

Several clinical phase compounds, including ABBV-744, Molibresib, and Pelapresib, are substrates for CYP3A4, indicating potential for extensive metabolism and significant drug-drug interactions. These interactions necessitate careful consideration when co-administering drugs that inhibit or induce CYP3A4, as they can alter the pharmacokinetics of these chemotherapeutics, impacting both efficacy and safety. CYP3A4 is inhibited by numerous compounds from the clinical phase group, while the guideline group compounds are still unaffected. CYP3A4 interactions can amplify therapeutic potency, but they can also lead to drug-drug interactions and potential toxicity ⁵⁸. A study reported shows that mitotane, a drug used to treat adrenocortical carcinoma, was found to induce the mRNA expression of CYP3A4 ⁵⁹. Moreover, some compounds exhibit broad inhibitory effects on multiple CYP enzymes. ABBV-744, for example, inhibits CYP1A2, CYP2C19, and CYP2C9, which can lead to increased plasma levels of other drugs metabolized by these enzymes, raising the risk of adverse effects. In contrast, guideline compounds like Daunorubicin show minimal inhibitory effects, suggesting fewer metabolic interactions but still requiring comprehensive pharmacokinetic assessment to ensure effective dosing.

The excretion and toxicity profiles reveal significant discrepancies. Drugs like Daunorubicin and Cyclophosphamide exhibit high risks of drug-induced liver injury (95.9% and 76.3% respectively) and cardiotoxicity. This is a critical finding, as pediatric patients have developing and fragile organ systems, being particularly vulnerable to long-term, irreversible systemic damage that can severely impact their quality of life as they grow older, as well as leaving them vulnerable for other health related issues. The predicted cardiotoxicity of drugs like Daunorubicin, for instance, is a well-documented clinical challenge that necessitates on the clock cardiac monitoring for these children. In contrast, clinical phase drugs such as ABBV-744 and Pelapresib show moderate half-lives and unique clearance profiles, with ABBV-744 displaying a significant total clearance rate (0.688 Log ml/min/kg) compared to SEL120’s much lower clearance (-0.167 Log ml/min/kg). In terms of half-life, which affects therapeutic efficacy and adverse impacts, only Daunorubicin and Vincristine in the control group have short durations, while three drugs from the clinical phase group have extended half-lives, suggesting longer drug effects or dosing modifications. In addition, a drug with a very long half-life, like most of the drugs in the guidelines, can lead to certain inappropriate adversities, such as delayed onset of rapid action when required ⁶⁰, difficulty in adjusting dosage due to the need for a longer period to reach stable concentrations ⁶¹, increased risk of accumulation, which can lead to a toxic level ⁶², increasing the risk of adverse events and life-threatening complications. Notably, in terms of younger patients, a long half-life coupled with lower clearance could increase the risk of toxic accumulation due to age-related variations in metabolic rates, and is a crucial side effect for guiding safer drug development for children.

Daunorubicin exhibits cytotoxic effects through various mechanisms. Studies have shown that daunorubicin induces double strand breaks (DSB) in ALL cell lines, leading to cell cycle arrest and reactive oxygen species generation. Genetic determinants play a role in the sensitivity to daunorubicin-induced cytotoxicity, with specific genes and pathways influencing the response to treatment ^63–65. Studies have demonstrated the hepatotoxic and haematological toxicities linked to 6-Mercaptopurine during chemotherapy for paediatric patients with acute lymphoblastic leukaemia, highlighting the necessity of dose modifications to control toxic effects ⁶⁶. Additionally, research suggests that 6-Mercaptopurine induces cytotoxicity in hepatocytes through reactive oxygen species formation, mitochondrial, and lysosomal damage, leading to cell death ^67,68. Furthermore, our comparative analysis revealed significant variability in the physicochemical properties and ADMET profiles between guideline compounds and those in clinical phases, emphasizing the need for integrating computational methods in early drug development to optimize chemotherapeutic candidates for pediatric leukemia. This integration is crucial for predicting ADMET properties, optimizing chemical libraries, and prioritizing potential drug candidates to improve safety and efficacy in treating pediatric leukemia.

The hERG potassium channel is vital for cardiac repolarization, and its inhibitors, like Vincristine from the control group and Mivebresib, ABBV-744, Molibresib, RO6870810, Pelapresib, INCB057643, Flavopiridol, FN-1501 and Pinometostat from the clinical phase group, can trigger severe arrhythmias ⁶⁹. Given the pronounced differences in pharmacokinetic profiles, it’s clear that compounds in the clinical phase group have a higher propensity to interact with pivotal drug transporters, proteins, and enzymes, emphasizing the need for thorough safety evaluations in clinical trials, especially considering the pronounced P-gp and hERG II inhibitions noted. In addition, p-glycoprotein is an efflux pump responsible for transporting drugs out of cells ⁷⁰. Its inhibitors, such as Mivebresib, ABBV-744, Molibresib, RO6870810, Pelapresib, and Pinometostat from the clinical phase group, can elevate intracellular drug concentrations, potentially improving efficacy but also heightening toxicity risks; in comparison, only Vincristine from the control group showed this property.

It is important to mention that the random forest model, due to the categorization of variables, does not provide adequate sizing, but it is useful to provide insights into the impact of properties (Fig. 4). The model indicates that the main differences between the two groups are related to P-gp I inhibition properties, h-ERG II inhibition, plasma protein binding rate, half-life and hepatotoxicity (Table 3). In general, these more divergent properties may indicate greater attention in terms of drug-drug interaction (P-gp I ⁷¹, volume of distribution (plasma protein binding rate) (PMID: 31,424,864), dose-concentration relationships (PK) and concentration-effect relationships (PD) (plasma protein binding rate) ⁷², cardiotoxicity (h-ERG II inhibition)⁷³, excretion rates as well as steady-state concentrations for a drug (half-life) (PMID: 32,119,385).

Table 3.

A comparison of pharmacokinetic properties between two chemotherapy drug groups revealed significant results for P-gp I inhibitor, hERG II inhibition, PPB, and half-life time. The drugs with positive values were identified.

Divergent properties	Guidelines g	Clinical phase g
P-gp I inhibitor	Vincristine	Mivebresib, ABBV-744, Molibresib, RO6870810, Pelapresib, Pinometostat
Plasma protein binding (PPB)	Daunorubicin	Mivebresib, ABBV-744, Pelapresib, Flavopiridol, BTX-A51
Half life	Citarabine, 6-Mercaptopurine, 6-Thioguanine, Cyclophosphamide, Prednisone, Dexamethasone, Methotrexate, Leucovorin	6-Mercaptopurine, Pracinostat, Pinometostat
hERG II inhibition	Vincristine	Mivebresib, ABBV-744, Molibresib, RO6870810, Pelapresib, INCB057643, Flavopiridol, FN-1501, Pinometostat

Optimization and energetic characterization

Computational analysis of the quantum properties of pharmaceutical compounds is crucial to predict their molecular behavior and seek to optimize their effectiveness. The tables summarizing the conformational analyzes, in addition to those of energies and quantum descriptors (Tables S1, S2 and S3) provide valuable information about the molecular characteristics of these compounds.

Regarding conformational analyzes, Daunorubicin exhibits 21,209 unique pharmacophore fingerprints and 64 conformations, with an average radius of gyration of 5.00 and a solvent accessible surface area of 742.80. Citarabine has 3,444 unique fingerprints and 52 conformations, with an average radius of gyration of 3.32 and a surface area of 426.73. Both 6-Mercaptopurine and 6-Thioguanine have simpler profiles, each with 12 and 60 unique fingerprints, respectively, and a single conformation, showing radii of gyration of 2.36 and 2.64 and surface areas of 299.13 and 322.09. Cyclophosphamide and Vincristine exhibit 347 and 36,842 unique fingerprints and 56 and 51 conformations, with average radii of gyration of 2.99 and 5.42 and surface areas of 421.19 and 993.87, respectively. Prednisone and Dexamethasone, with 3,759 and 5,750 unique fingerprints and 31 and 21 conformations, have average radii of gyration of 3.78 and 3.88 and surface areas of 547.40 and 566.52. Methotrexate and Leucovorin, with 29,383 and 37,401 unique fingerprints and 254 and 87 conformations, have average radii of gyration of 5.27 and 5.23 and surface areas of 715.91 and 719.39, respectively (Table S4).

Newer clinical phase compounds also exhibit diverse profiles. Mivebresib, ABBV-744, and Molibresib have 4,195, 11,290, and 5,992 unique fingerprints and 87, 168, and 109 conformations, with radii of gyration of 4.55, 4.85, and 4.33 and surface areas of 672.00, 764.20, and 675.87, respectively. RO6870810 and Pelapresib, with 16,217 and 2,782 unique fingerprints and 183 and 25 conformations, show radii of gyration of 5.17 and 3.90 and surface areas of 795.43 and 586.08. INCB057643, CYC065, and Flavopiridol present 1,920, 12,191, and 6,697 unique fingerprints and 16, 200, and 77 conformations, with radii of gyration of 4.22, 4.68, and 4.15 and surface areas of 634.70, 683.17, and 604.37. FN-1501, BTX-A51, FLX925, SEL120, Birabresib, AMG900, Pracinostat, and Pinometostat show considerable diversity with unique fingerprints ranging from 5,850 to 34,493 and conformations from 12 to 254 (Table S4).

Quantum calculations of chemotherapeutics compounds

Energetic characterization

The analysis of the energetic and structural properties of compounds listed in both the guidelines and clinical phases (I and II) reveals significant differences that may influence their therapeutic applications. Based on total and binding energies, compounds in the guidelines, such as Vincristine, Methotrexate, and Leucovorin, demonstrate high binding energies, indicating greater stability in the context of specific molecular interactions. Specifically, Vincristine exhibits the highest binding energy (-192.6647269 Ha) and a considerable surface area (2259.210822 Ų), suggesting strong interaction potential and stability in biological environments. This high binding energy and large surface area imply that Vincristine might form more stable and extensive interactions with its target, which could translate to higher efficacy in clinical applications ^74–76 (Table S5).

Methotrexate and Leucovorin also exhibit substantial binding energies, -106.5107569 Ha and -112.2119875 Ha respectively, combined with significant surface areas and cavity volumes, indicating that these compounds might interact effectively with biological targets. Methotrexate’s relatively high dipole moment (2.17391978 Debye) suggests a strong polarity, which could enhance its interaction with polar biological molecules or environments. Similarly, Leucovorin’s high dipole moment (3.92272941 Debye) supports its potential for significant interactions in biological systems. The solvation energies of these compounds, while lower than their binding energies, are still notable and imply reasonable solubility and bioavailability (Table S5).

The larger total energies of 6-Mercapt and 6-Thiog (-848.0839406 and -906.3346865 Ha) indicate less stability. Additionally, they exhibit high binding energies (-42.09272673 and -45.4116311 Ha), indicating weaker binding interactions. With low dipole moments (3.19289546 and 3.51983108), 6-Mercapt and 6-Thiog demonstrate low solubility and reduced interaction with polar environments, potentially benefiting their distribution in blood plasma. Their lower dielectric solvation energies (-0.04352734 and -0.05172271 Ha) further confirm reduced solubility. 6-Mercapt and 6-Thiog have smaller surface areas (574.2578458 and 622.7498642 Ų) and lower cavity volumes (1043.202967 and 1130.653467 ų), indicating a potentially less efficient fit within the active sites of enzymes or receptors (Table S5). These findings corroborate experimental data, where the oral bioavailability of 6-MP is low (16%–50%) due to its insolubility (0.22 mg/mL) and its short half-life (0.5–1.5 h) ⁷⁷. For this reason, our analyzes are crucial for the development of effective and safe drugs. These computations offer comprehensive insights for various strategies, including those based on nanotechnology ⁷⁸.

Comparatively, compounds in clinical phases, such as SEL120 and RO6870810, also exhibit high binding energies, with SEL120 showing a binding energy of -164.7671702 Ha and RO6870810 showing -137.7870018 Ha. These compounds also present substantial dipole moments (3.57212689 Debye for SEL120 and 7.20248601 Debye for RO6870810) and surface areas (1144.260853 Ų for SEL120 and 1693.605637 Ų for RO6870810), indicating potential for robust interactions. The high dipole moments suggest that these compounds are highly polar, which might enhance their interactions with polar environments or molecules in the body (Table S5).

However, their solvation energies and dielectric solvation energies are generally lower than those of the guideline compounds, which could affect their solubility and distribution in biological systems. For instance, SEL120 and RO6870810 have dielectric solvation energies of -0.02659583 Ha and -0.06393046 Ha, respectively, compared to -0.08128777 Ha for Methotrexate. This lower solvation energy could mean that the clinical phase compounds are less soluble in aqueous environments, potentially affecting their bioavailability and efficacy. Additionally, some clinical phase compounds such as Molibresib and ABBV-744 show exceptionally high dipole moments (7.35663455 Debye and 3.5298378 Debye respectively), which might suggest strong polar interactions. However, their binding energies are slightly lower than some guideline compounds, indicating a potential trade-off between binding affinity and solubility (Table S5).

Quantum chemical descriptors

Quantum chemical descriptors provide a more intrinsic view and deeper reasoning for the observed pharmacological profiles, providing a better background for some pharmacokinetic behaviors observed. In our study, compounds such as Cyclophosphamide stood out with a high HOMO–LUMO gap (-7.51 eV), indicating remarkable chemical stability. From a clinical perspective, however, this high stability is a double-edged sword: it often correlates with a predictable pharmacokinetic profile and a well-documented, manageable toxicity profile, and as such these drugs have been used as go-to therapeutic options for decades; however, this stability may also be associated with lower intrinsic reactivity at the binding site and worse interacting profiles to the target molecules. Conversely, Molibresib has a smaller GAP, higher electrophilicity (20.98 eV), and high chemical potential (4.27 eV), suggesting greater chemical reactivity. This increased reactivity may translate into higher target binding affinity and greater clinical efficacy, which are the therapeutic hypotheses being tested in Phase I/II trials. On the other hand, it may also increase the risk of off-target interactions or production of toxic metabolites, which may lead to unforeseen side effects. Therefore, our quantum analysis not only describes molecular properties, but rather quantifies the fundamental trade-off between the proven, predictable nature of established agents and the potentially higher efficacy — but also higher risk—of novel therapeutics. (Table S6).

Prednisone, Dexamethasone, and Cyclophosphamide demonstrate a combination of high reactivity, stability, and desirable chemical properties. Prednisone exhibits a high electrophilicity index (23.36 eV) and electronegativity (4.34 eV), indicating strong reactivity that can enhance therapeutic efficacy. It also has a significant ionization potential (6.82 eV) and chemical hardness (2.48 eV), reflecting its stability, making it effective for interacting with biological targets while maintaining a stable profile, thereby reducing potential side effects in pediatric patients. Dexamethasone shows similar high electrophilicity (23.65 eV) and electronegativity (4.4 eV), respectively, implying robust reactivity. With an ionization potential of 6.84 eV and a chemical hardness of 2.45 eV, it also promises good stability, making it a strong candidate for pediatric chemotherapy, balancing efficacy with safety. Cyclophosphamide stands out with the highest GAP value (-7.51 eV), indicating exceptional chemical stability crucial for minimizing toxicity and adverse effects in pediatric patients. Its notable electrophilicity index (20.61 eV), high ionization potential (7.07 eV), and chemical hardness (3.76 eV), respectively, further underscore its potential as a stable and effective chemotherapeutic (Table S6).

The table of compounds in clinical phases (I and II) reveals a range of values for each parameter. The HOMO values range from -6.59 eV (Pelapresib) to -5.33 eV (Pinometostat), while the LUMO values range from -1.97 eV (Molibresib and Flavopiridol) to -0.48 eV (SEL120). The GAP values span from -4.96 eV (CYC065) to -3.74 eV (AMG900). For ionization potential (I), the highest value is 6.59 eV (Pelapresib) and the lowest is 5.33 eV (Pinometostat). Electron affinity (A) varies between 1.97 eV (Molibresib and Flavopiridol) and 0.48 eV (SEL120). Chemical hardness (η) ranges from 2.48 eV (CYC065) to 1.87 eV (AMG900), while softness (σ) varies from 0.54 (AMG900) to 0.4 (CYC065 and SEL120). Chemical potential (μ) spans from 4.27 eV (Molibresib) to 2.96 eV (SEL120). Electronegativity (χ) ranges from -4.27 eV (Molibresib) to -2.96 eV (SEL120), and the electrophilicity index (ω) spans from 21.14 eV (Pelapresib) to 10.68 eV (Pinometostat) (Table S6).

Based on the analysis of the compounds in clinical phases (I and II), Pelapresib and Molibresib stand out as particularly promising candidates. Pelapresib demonstrates the highest GAP value (-4.72 eV), indicative of good stability, combined with the highest ionization potential (6.59 eV) and the highest electrophilicity index (21.14 eV), suggesting strong reactivity and potential efficacy in therapeutic applications. Molibresib also shows a high electrophilicity index (20.98 eV) and chemical potential (4.27 eV), indicating robust reactivity and effective interaction with biological targets. Additionally, Molibresib’s combination of high ionization potential (6.57 eV) and significant electron affinity (1.97 eV) reflects its balanced stability and reactivity, making it a interesting candidate for further priorization and investigation in pediatric chemotherapy (Table S6).

Limitations

This study is entirely in silico and was designed to be comparative and hypothesis-generating, rather than to provide definitive quantitative predictions of in vivo behavior. ADMET descriptors from different servers and DFT-based quantum parameters may vary according to underlying algorithms and static geometric assumptions, to mitigate this, we focused on cross-platform consistency and global trends rather than on absolute values for individual compounds. Although experimental validation and dynamic modeling were beyond the scope of this work, our integrated framework still offers a structured map of predicted pharmacokinetic and reactivity profiles that can guide subsequent experimental studies and is readily extendable beyond pediatric ALL to other oncologic contexts.

Conclusion

This study offers a comprehensive comparative analysis of chemotherapeutic agents from the 2009 BFM and GBTLI guidelines with those currently in clinical phases I and II. Utilizing computational methods, we assessed drug similarity, predicted ADMET properties, and conducted quantum mechanical calculations using DFT. The results demonstrated that certain compounds, including RO6870810 and Pinometostat, displayed unfavorable values for specific metrics. In contrast, other clinical-phase compounds, such as Pelabresib and Molibresib, exhibited favorable drug-likeness properties, with ADMET profiles, high electrophilicity, and quantum descriptors indicative of strong target engagement potential, while maintaining a balance between reactivity and stability. These properties suggest that they may offer therapeutic advantages over traditional cytotoxic agents. Given their progression into phase I/II trials, our in silico findings further support prioritizing Pelabresib and Molibresib for continued preclinical optimization and clinical evaluation as next-generation therapies for pediatric ALL. Although the present work focuses on pediatric ALL, the integrated pharmacoinformatic and quantum-descriptor workflow is generalizable and can be applied to other oncologic indications to compare established treatments with investigational compounds within a unified computational framework.

For pediatric oncology, these findings help to prioritize promising candidates and to rationalize the optimization of existing therapies. While in silico assays provide valuable preliminary information, clinical validation is essential. This study underscores the value of applying computational methods early in drug discovery to anticipate pharmacological and toxicological profiles and to guide candidate selection and optimization. Nonetheless, systematic in vitro and clinical studies are required to validate and refine these predictions. As with all in silico approaches, our findings represent approximations of in vivo behavior supported by statistical models. Continued research combining advanced computational techniques with experimental validation will be essential to develop safer and more effective therapies for pediatric cancer. To translate these insights, a focused validation path is required, beginning with targeted in vitro assays for key predictions (e.g., transporter and ion-channel interactions) and progressing to preclinical models to confirm pharmacokinetic behavior and antileukemic activity, thereby integrating computational insights with experimental drug development.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Ian A. F. Bahia contributed to conceptualization, methodology, and original draft preparation. Maria K. da Silva contributed to data curation, investigation, and formal analysis. Emad Rashad Sindi contributed to methodology and validation. João F. Rodrigues-Neto contributed to software, data analysis, and visualization. Edilson D. da Silva Jr. contributed to investigation and resources. Taha Alqahtani contributed to validation and critical review of the manuscript. Yewulsew Kebede Tiruneh contributed to supervision, project administration, and critical revision of the manuscript. Magdi E. A. Zaki contributed to supervision, funding acquisition, and final approval of the manuscript. Umberto L. Fulco contributed to conceptualization, methodology, and manuscript review. Jonas I. N. Oliveira contributed to formal analysis, writing—review and editing. All authors read and approved the final manuscript.

Funding

The authors extend their appreciation to the Deanship of Scientific Research at King Khalid University for funding this work through large group Research Project under grant number RGP2/413/46.

Data availability

Data are available to the corresponding author upon reasonable request.

Declarations

Competing interests

On behalf of all authors, the corresponding author declares that there is no conflict of interest.

Informed consent

Not applicable.