D2Screen: Embedding Pretrained Representation Learning Model and Molecular Docking for Virtual Screening
Drug discovery is notoriously resource-intensive, prompting interest in computational methods such as virtual screening. Traditional approaches to virtual screening, viz., receptor-based methods (such as molecular docking) and ligand-based methods, face limitations with respect to accuracy and scalability. Recently, integrating artificial intelligence (AI) with molecular docking has emerged as a powerful way to explore vast chemical libraries more efficiently. AI models capture complex ligand–protein interactions but often lack interpretability, whereas molecular docking provides structural insights but struggles with highly flexible or dynamic targets.
Appearing in this issue as a Featured Letter, Qian et al. (DOI: 10.1021/acsmedchemlett.5c00433) introduce D2Screen, a hybrid virtual screening pipeline that integrates a pretrained molecular representation model (GEM) with molecular docking, combining the strengths of ligand- and structure-based strategies. In the first stage, GEM, which can be fine-tuned with target-specific bioactivity data, rapidly screens millions of compounds for potential activity. In the second stage, promising candidates are subsequently evaluated by molecular docking, which supplies interpretable binding poses and ranked binding affinity scores. Benchmarking revealed that D2Screen consistently outperformed either AI or docking alone, particularly for challenging targets such as GPCRs and cytochrome P450 enzymes. The approach also proved robust across variations in data set size, activity ratio and scaffold diversity.
A case study targeting the SARS-CoV-2 main
protease (Mpro) demonstrated D2Screen’s practical
value. The pipeline identified
novel noncovalent inhibitors, including quinoline-based compounds
that retained activity against drug-resistant Mpro mutants
(notably E166 V) known to undermine therapies such as nirmatrelvir.
Structural analysis and molecular dynamics confirmed reduced reliance
on mutation-prone residues, offering a rational basis for designing
resistance-evading inhibitors. Overall, D2Screen represents a scalable,
modular and effective framework for AI-driven drug discovery, delivering
improved enrichment, improved accuracy and resilience to resistance
challenges.
Development of Purine and Pyrrolopyrimidine Scaffolds as Potent, Selective, and Brain Penetrant NUAK1 Inhibitors
NUAK1 (also known as ARK5) is a serine/threonine kinase within the AMPK-related kinase (ARK) family and is implicated in oncogenesis, neurodevelopment, and neurodegeneration. Dysregulated NUAK1 promotes tumor cell adhesion, migration and survival, while also contributing to neuronal development and tauopathies such as Alzheimer’s disease. This dual role in cancer and brain disorders has driven interest in developing selective inhibitors with favorable pharmacokinetic and brain penetration profiles. Earlier inhibitors of NUAK1 lacked selectivity, brain exposure or metabolic stability. The CDK2 inhibitor NU6140 demonstrated potent activity against NUAK1 but suffered from poor metabolic stability and highly efflux, limiting its utility in the CNS.
In this issue, Aldred et al. (DOI: 10.1021/acsmedchemlett.5c00467) report efforts to optimize NU6140 derivatives. Purine-based analogs retained potency but often showed poor metabolic stability or off-target activity against CDKs. Substituting the cyclohexyl group with a tetrahydropyran (compound 20, ARUK2010694) yielded improved metabolic stability, peripheral selectivity and in vivo half-life, though brain exposure remained low. To achieve CNS penetration, the team explored pyrrolopyrimidine analogs. This scaffold lowered polar surface area and efflux, thereby enabling higher brain exposure. Compound 23 (ARUK2010489) emerged as a potent, selective NUAK1 inhibitor with significant CNS penetration.
Overall, this study developed
two complementary tool compounds:
ARUK2010694 (20), a selective peripheral NUAK1 inhibitor,
and ARUK2010489 (23), a brain-penetrant inhibitor suitable
for CNS studies. These molecules provide critical chemical probes
for investigating NUAK1 biology in cancer and neurodegeneration.
Unraveling the Potential Pathogenic Role of Squalene Synthase (SQS) in Lung Cancer Using Enzyme Inhibitors as Molecular Tools
Squalene synthase (SQS) catalyzes the first committed step of cholesterol biosynthesis by diverting farnesyl pyrophosphate toward sterol production. By regulating cholesterol and sterol derivatives critical for membrane integrity, lipid raft formation and signaling, SQS plays a central role in normal physiology. In cancer, however, dysregulated cholesterol metabolism promotes proliferation, survival and metastasis. Elevated SQS expression is particularly evident in nonsmall cell lung cancer, where it drives invasion and poor prognosis through pathways including EGFR, PI3K/Akt, Ras/MAPK, and NF-κB/MMP1. In addition, squalene accumulation exerts cytoprotective effects that can be reversed by the inhibition of SQS, thereby promoting oxidative stress and ferroptotic cell death in cancer cells. Although SQS inhibitors were originally developed for hypercholesterolemia, they were abandoned due to toxicity or limited efficacy compared to statins. Their anticancer potential, however, is now gaining renewed attention.
In their recent work,
Katavati et al. (DOI: 10.1021/acsmedchemlett.5c00458) sought to provide evidence of SQS as a potential druggable target
in lung cancer. Two chemically distinct SQS inhibitors (compound 1, a propylamine derivative, and compound 2,
a morpholine analog) were used as pharmacological tools in murine
and human lung cancer cell lines. Both compounds demonstrated potent,
selective cytotoxicity against highly metastatic cells while sparing
normal fibroblasts. Mechanistic assays revealed that they impaired
adhesion, induced cell-cycle arrest (G0/G1 or G2), triggered early
apoptosis and promoted mitochondrial hyperpolarization. In vivo, compound 1 markedly reduced lung metastases in a mouse model, particularly
under preventive treatment. These findings position SQS as a promising
and druggable target in lung cancer. Future work should focus on optimizing
inhibitor design, delivery strategies and combination regimens to
fully harness the potential of SQS inhibition in treating both primary
and metastatic disease.