Alexey V. Zakharov

Predicting compounds that are more potent than those represented in the training data remains a difficult and practically important problem for quantitative structure-activity relationship (QSAR) modeling. This challenge is particularly relevant during lead optimization, where the goal is to design or identify analogs that are more potent than compounds already measured. To address this problem, we developed a new Pairwise-Difference QSAR (PD-QSAR) approach that learns changes in biological activity from differences in molecular descriptors between compound pairs. We evaluated three strategies for aggregating the resulting PD-QSAR predictions: mean aggregation, k-nearest-neighbor aggregation (k = 3), and 10% quantile aggregation, and compared their performance with Direct QSAR and Chemprop using a benchmark derived from publicly available ChEMBL 35 data. The primary benchmark included 103 human single-protein target datasets, comprising 33,707 activity records. For each target, the 10% most potent activity records were withheld for external validation. Among the approaches evaluated, 10% quantile aggregation provided the lowest median external validation RMSE (0.518) and outperformed mean and kNN aggregation on external validation RMSE for 100 of 103 targets, Direct QSAR for 102 targets, and Chemprop for 94 targets. These results support PD-QSAR with quantile aggregation as a useful strategy for prioritizing analogs expected to extend beyond the potency range represented by previously measured compounds.

Filovirus outbreaks caused by Ebola virus (EBOV) and Marburg virus (MARV), pose severe global health threats characterized by high rates of fatal hemorrhagic fever. While species-specific vaccines and therapeutic monoclonal antibodies are approved for Zaire ebolavirus, broadly-active therapeutics remain unavailable, leaving populations vulnerable to MARV and other pathogenic Ebola species, such as Bundibugyo (BDBV) and Sudan (SUDV) ebolaviruses. Here we report a computationally guided, infectious virus validated screening platform for the rapid discovery of broad-spectrum filovirus antivirals. By leveraging quantitative structure–activity relationship (QSAR) models, we screened 142,382 compounds in silico to prioritize 125 high-potential candidates. Subsequent dose–response and viability profiling identified 23 compounds exhibiting potent, low-micromolar pan-filovirus activity and favorable cytotoxicity profiles. Molecular docking indicates these compounds target conserved structural and functional domains—primarily the VP35 and L proteins—which may disrupt essential viral replication and immune antagonism. Furthermore, systematic combinatorial screening revealed three highly synergistic compound pairs, notably NCGC00113249-01 and NCGC00118008-01, demonstrating robust cross-species efficacy. By targeting conserved vulnerabilities across the filovirus family, this integrated in silico and in vitro pipeline provides a scalable framework to rapidly nominate and optimize synergistic therapeutic regimens against both endemic and emerging viral threats including BDBV. Graphical

Developing potent, selective small-molecule inhibitors remains a major challenge in drug discovery. ALDH3A1, a detoxifying aldehyde dehydrogenase isoform implicated in cancer and neurodegeneration, is a promising yet underexplored therapeutic target. To accelerate inhibitor optimization, we developed an AI-guided, reaction-based hit-to-lead workflow combining sequential reaction enumeration, pharmacophore-informed docking, and predictive modeling to support scalable SAR expansion. Applied to ALDH3A1, two rounds of enumeration using Enamine building blocks generated about 250,000 virtual analogues. Combined deep learning and docking-based triage prioritized 150 compounds for synthesis, leading to a roughly 1,000-fold improvement in biochemical potency from 1.41 μM to 1 nM for NCATS-SM0707, together with 4 nM cellular activity for NCATS-SM0708. Crucially, this methodology can be expanded by applying various chemical reactions at different positions. This study highlights CSAR as a scalable, generalizable complementary strategy to accelerate hit optimization through reaction-based enumeration and AI-guided prioritization of larger synthetically accessible chemical space.

Endoplasmic reticulum-associated degradation (ERAD) is a critical protein quality control mechanism that also regulates lipid metabolism and calcium homeostasis. Dysregulation of ERAD and unfolded protein response underlies diseases including cancer, neurodegenerative disorders, and metabolic syndromes. Small molecule modulators of ERAD could enable mechanistic discovery and therapeutic intervention, but few have been identified. Using a high-content screening, we discovered several ERAD-modulating compounds, including NCATS-SM0225, an ERAD inhibitor that unexpectedly binds all three isoforms of VDAC, outer mitochondrial membrane proteins enriched at mitochondria-associated membranes. This led us to discover an essential role for VDACs in ERAD and ER-phagy. NCATS-SM0225 elevates cytosolic, ER, and mitochondrial calcium through calcium influx and IP3R–MCU activity. This calcium imbalance strengthens VDAC1–IP3R coupling and activates PERK, which phosphorylates STIM1 and drives degradation of key ERAD regulators. Loss of these components amplifies PERK signaling and selectively kills cancer cells while sparing normal cells. These findings uncover a cancer-specific role of VDACs in ERAD regulation and calcium signaling, highlighting a therapeutically actionable vulnerability.

Rapid advances in biomedical research and nanomedicine have driven the development and implementation of New Approach Methodologies (NAMs) for predictive evaluations of novel formulations.

Nucleic acid nanoparticles (NANPs) represent a versatile platform for drug delivery and modulation of therapeutic responses. To expedite NANPs’ translation from bench to bedside, rapid coordination of their design principles with immunostimulatory assessment is essential. Here, a deep learning framework is presented to predict cytokine responses, specifically interferon‐beta (IFN‐β) and interleukin‐6 (IL‐6), induced by NANPs in human microglial cells based solely on their sequences. Using a transformer‐based architecture augmented through systematic strand permutation trained on 176 structurally diverse, individually assembled, and experimentally characterized NANPs, the model achieved high predictive performance in cross‐validation (R ² = 0.96–0.97, RMSE ≤ 0.01) and demonstrated strong generalizability on an external test set (R ² = 0.91 for IFN‐β; 0.85 for IL‐6). This work advances sequence‐based quantitative structure‐activity relationship (QSAR) modeling by leveraging attention‐based neural networks to eliminate the need for manual feature engineering while maintaining biological interpretability. To facilitate community access, the updated artificial immune cell (AI‐cell) web‐based platform is introduced, which supports rapid immune profiling of NANPs in silico . This new approach methodology provides a scalable framework to guide the rational design and optimization of NANPs through rapid prediction of their immune responses.

Cytochrome P450 (CYP) family of enzymes plays an integral role in drug metabolism and excretion. This application note describes the development of a novel computational CYP profiler (CYP-Pro) as a drug development tool. To enable new model development, we integrated and curated the largest, to the best of our knowledge, dataset comprising 26587 entries, including both inhibitors and substrates of CYP2D6, CYP3A4, and CYP2C9. We have built and externally validated Quantitative Structure-Activity Relationship (QSAR) models that can accurately predict whether molecules of interest are expected to be inhibitors or substrates. The models were assessed mainly by Positive Predictive Value (PPV), which ranged between 0.14 and 0.92. CYP-Pro showed the highest accuracy in predicting compounds selectively metabolized by CYP3A4 alone or by CYP2D6 and CYP2C9 without CYP3A4 involvement. All models are incorporated into the previously developed PhaKinPro portal (https://phakinpro.mml.unc.edu). CYP-Pro is unique in that it provides separate models for predicting CYP inhibitors vs. substrates, prioritizes high positive predictive value (PPV) as a pragmatic metric of accuracy to support the experimental testing of a small number of predicted substrates and inhibitors, enhances interpretability with fragment maps, and ensures reliability through strict applicability domain control. We expect that this new tool will aid researchers in early identification of compounds with favorable metabolic profiles, reducing the risks of drug-drug interactions and improving the efficiency of drug development efforts.