💻 Computational & AI

AI drug discovery, cheminformatics, bioinformatics, molecular simulation and virtual screening.

2026-10-23 1 article

2026-10-21 · 💻Computational chemistry / neural network potential for ligand strain · Journal of Chemical Information and Modeling

A solvated neural-network potential makes ligand strain estimates usable at library scale

AI & machine learning Molecular simulation

Conventional strain estimates reference a single gas-phase minimum, which ignores solvent and distorts charged ligands whose internal electrostatics collapse them into compact structures in vacuo.
AIMNet2-CPCM, a neural network potential trained on DFT-level energies and forces in a continuum solvent, handles neutral and charged species on equal footing.
GPU batch inference makes a solvated global-minimum search feasible across a full screening library.
Across roughly 8,000 PDB ligands, solvated strain rises with net charge, with a trend that holds up to a charge of −2.

💡 Points

Strain filters are only as good as the unbound reference, and anyone who has seen a zwitterion fold into a salt bridge in vacuo knows why gas-phase numbers mislead. A solvated, charge-aware potential fast enough for library triage is something I would actually put in a docking post-filter. The caveat is the less reliable behavior at higher negative charge, which matters for acids and phosphates.

Efficient and Accurate Ligand Strain Calculations in Solution with the AIMNet2 Neural Network Potential ↗

2026-10-22 1 article

2026-10-20 · 💻AI cofolding models / kinases and GPCRs · Journal of Chemical Information and Modeling (ACS)

Cofolding models can place the ligand right while getting the protein conformation wrong

Virtual screening & docking AI & machine learning Kinases GPCRs

AlphaFold3, RoseTTAFold3, Boltz-2 and Chai-1 were benchmarked on kinases and Class A GPCRs, families whose pharmacology depends on defined conformational states.
Only complexes released after all four models' training cutoffs were used, plus one earlier control structure, to give a prospective test.
Correct ligand placement turned out to be systematically decoupled from accurate prediction of the surrounding protein conformation.

💡 Points

For anyone docking into cofolded poses, this is the warning label: a plausible ligand pose does not mean the DFG state or GPCR activation state is right. That matters most for type II kinase inhibitors and biased GPCR ligands, where the conformation is the whole point. I'd treat cofolded structures as hypotheses to validate against known conformational markers before using them for SAR design.

Benchmarking Protein–Ligand Cofolding Models: Correct Ligand Placement Is Decoupled From Accurate Protein Conformation ↗

2026-10-21 1 article

2026-10-19 · 💻P-gp efflux / computational design · Journal of Chemical Information and Modeling (ACS)

Fragment-synergy genetic algorithm turns P-gp substrates into non-substrates

DMPK & ADME

P-gp efflux limits bioavailability and CNS exposure, and ML substrate classifiers rarely translate into concrete design moves.
SF-GA builds fragment synergy analysis into the mutation step of a genetic algorithm, biasing replacements toward substructures that co-occur in non-substrates.
Against GA and reinforcement-learning baselines at a fixed evaluation budget it was competitive and gave the most synthetically accessible molecules, reporting the removed and inserted fragments for every accepted mutation.

💡 Points

The part I would use is the per-mutation fragment report: it reads like a matched-pair suggestion list, which is how a chemist actually wants a model to talk. Competitive rather than better optimization performance means the value is interpretability and synthesizability, not raw efficiency. Prospective validation on real efflux assays is the missing piece.

Synergy-Guided Molecular Redesign of Predicted P-Glycoprotein Substrates toward Synthetically Feasible Non-Substrates ↗

2026-10-20 1 article

2026-10-16 · 💻Kinase inhibitors (protonation/tautomer modelling) · Journal of Chemical Information and Modeling (ACS)

Kinase inhibitors often change protonation or tautomer state on binding, MCCE modelling shows

Molecular simulation Kinases

Multi-conformation continuum electrostatics was applied to 37 co-crystal structures of nine kinase domains with 18 approved inhibitors.
Protein net charge stayed largely stable on binding, whereas inhibitor charge states frequently shifted up or down.
Binding can resolve two near-equal solution tautomers into one, creating free-energy differences of a few kcal/mol.

💡 Points

Docking with one fixed protonation state is quietly baked into most of our virtual screening, and this says it can be wrong by kcal/mol for approved drugs. When a scaffold with a tautomerisable heterocycle misbehaves in SAR, I'd now check its bound-state tautomer before blaming the pocket.

Kinase Inhibitors can Change Protonation or Tautomeric State upon Binding ↗

2026-10-19 1 article

2026-10-11 · 💻Covalent ligand docking (computational tool) · Journal of Chemical Information and Modeling (ACS)

CHARMM-GUI adds a web module that automates covalent ligand docking

Covalent drugs Virtual screening & docking

CGUI-CLD is a new CHARMM-GUI module that automates ligand preparation, ligand modification and docking runs for covalent ligands, built on AutoDock4.
A knowledge-based library covers 66 warheads and 8 amino acids, so the post-reaction ligand geometry no longer has to be built by hand.
Results are presented through a browser interface.

💡 Points

Covalent docking is where my own campaigns stall: building the bonded adduct for each warhead is tedious, so I triage by intuition. A browser tool that handles it would let me rank a whole warhead set against a cysteine pocket in an afternoon. Docking scores won't capture reactivity, so I'd still pair this with kinetic data.

CHARMM-GUI Covalent Ligand Docker as a Web-Based Molecular Docking Platform for Covalent Ligands ↗

2026-10-18 1 article

2026-10-16 · 💻Computational lead optimization (QSAR) · chemRxiv

Pairwise-difference QSAR with quantile aggregation extrapolates beyond known potency

SAR & lead optimization AI & machine learningCheminformatics

Standard QSAR struggles to predict analogs more potent than anything in its training set, which is exactly the lead-optimization question.
Pairwise-Difference QSAR learns activity changes from descriptor differences between compound pairs, then aggregates the predictions.
On ChEMBL-derived benchmarks across about a hundred targets with the most potent compounds held out, a low-quantile aggregation beat mean and kNN aggregation, direct QSAR and Chemprop.

💡 Points

Every optimization team has felt this: the model ranks the series fine, then fails at the top end where you actually need it. Holding out the most potent decile is the right stress test. My caution is that public ChEMBL tails differ from a live program's SAR cliffs, so I'd trial it prospectively on one of our own series before trusting it to prioritize synthesis.

Breaking the Potency Barrier: A Practical Recipe for QSAR Extrapolation ↗

2026-10-15 1 article

2026-09-18 · 💻DNA-encoded libraries (ML / docking / cofolding benchmark) · Journal of Chemical Information and Modeling (ACS)2026, 66 (19), 12360

Benchmark asks whether structure-based methods rescue DEL-trained ML out of distribution

DNA-encoded libraries Virtual screening & docking AI & machine learningCheminformatics

A recent NeurIPS competition showed top ML models trained on DEL data failing to generalize to out-of-distribution chemical space; this study tests whether structural modeling can bridge the gap.
It benchmarks ML, Glide docking, GALigandDock and Boltz-2 cofolding on three biologically diverse targets screened against DELs built in several synthesis formats.
ML does well in-distribution; the paper's focus is how each approach holds up on new chemistry.

💡 Points

DEL-trained models that shine on random splits and collapse on new chemotypes is exactly what bites when you try to seed a different series from a DEL campaign. Three targets is a small panel, but a like-for-like comparison with physics-based methods on identical data is useful. I'd use it to calibrate how much to trust ML triage on scaffolds the library never sampled.

Assessing the Generalizability of Machine Learning and Physics-Based Methods with DNA-Encoded Libraries ↗

2026-10-12 1 article

2026-10-10 · 💻Protein–protein binding energy prediction (incl. intrinsically disordered partners) · Journal of Chemical Information and Modeling (ACS)

Interface graph neural network predicts binding energy for folded and disordered partners

Virtual screening & docking AI & machine learning

FlexENN is a message-passing graph neural network built on a local interface graph with geometry, flexibility descriptors and sequence per residue.
Two models are trained independently: one for folded complexes and one for complexes with a disordered partner, using a newly built IDP dataset.
Benchmarks include tubulin–tubulin interfaces in microtubules and microtubule–MAP complexes.

💡 Points

Affinity predictors trained on folded complexes usually fall apart on disordered partners, so a dedicated IDP set is the real contribution here. For med chem it is protein–protein interface work rather than small-molecule scoring, and I'd test whether it ranks mutations or peptide variants reliably.

FlexENN: A Graph Neural Network for Binding Energy Prediction of Globular and Intrinsically Disordered Protein Complexes ↗

2026-10-10 1 article

2026-10-08 · 💻Computational fragment-based ligand design · Journal of Chemical Information and Modeling (ACS)

FragDockRL grows synthesizable ligands from a docked core with reinforcement learning

Fragment-based drug discovery Virtual screening & docking

FragDock builds molecules from synthetically accessible building blocks via known reactions and scores them by tethered docking that restrains a predefined core pose.
FragDockRL adds reinforcement learning: rewards from rDock tethered docking followed by SMINA scoring drive a modified Deep Q-Network that guides stepwise molecular growth.
The method was evaluated on three therapeutically important protein targets, including a receptor tyrosine kinase.

💡 Points

Growing from a fixed core under synthesis constraints mirrors how fragment elaboration actually happens on a project, which beats free generative models that hand you unmakeable structures. Docking-based rewards remain the weak link because they reward size, so I'd judge it by whether the compounds are truly makeable and hold up against a prospective test.

FragDockRL: A Reinforcement Learning Method for Fragment-Based Ligand Design via Building-Block Assembly and Tethered Docking ↗

2026-10-08 1 article

2026-10-06 · 💻Dengue virus NS4B inhibitors · Journal of Chemical Information and Modeling (ACS)

Modeling-led design yields a pan-serotype dengue NS4B inhibitor candidate

Structure-based designCheminformaticsInfectious disease

Comparative models of the structurally unresolved NS4B across all four dengue serotypes reveal a conserved transmembrane pocket that fits known resistance mutations.
Docking of the JNJ-series inhibitors fed predictive field-based 3D-QSAR models, which were used to screen bioisosteres and in-house compounds.
The top candidate, BNJ-495, was extremely potent against serotype 2 with a high selectivity index, and its conserved binding contacts suggest broad-spectrum activity.

💡 Points

A homology-model plus 3D-QSAR workflow that delivers a prospectively validated hit against a membrane protein with no structure is rare, and that is what makes this worth a look. Before I trust 'pan-serotype', I'd want to see the other three serotypes tested and whether the JNJ-series liabilities carried over.

Structure-Guided Design of a Subnanomolar Dengue Virus NS4B Inhibitor ↗

2026-10-04 1 article

2026-10-02 · 💻Protein–ligand scoring (virtual screening) · Journal of Chemical Information and Modeling (ACS)

DESPOT adds directionality to knowledge-based protein–ligand scoring

Virtual screening & docking Molecular simulation

Knowledge-based potentials are interpretable scoring functions, but most treat space around protein atoms as isotropic.
DESPOT is an all-atom anisotropic potential that assigns atoms to isotropic, axially symmetric or fully anisotropic classes by hybridization and bonding.
It uses a positionally averaged reference state and a 'void' ligand atom type so regions preferentially left empty can be represented.
The authors position it as a direction-aware successor to existing knowledge-based scoring.

💡 Points

Directionality is exactly what hydrogen bonds, halogen bonds and pi contacts need and what isotropic potentials blur, so this targets a real weakness in how I triage docked poses. The void atom type is a neat touch for pocket regions that ligands avoid. What I will check before trusting it: performance on my own targets and decoy sets, not just benchmarks, and whether it ranks congeneric series rather than only separating actives from decoys.

DESPOT: Direction-Enhanced Scoring POTentials for Scoring of Protein–Ligand Interactions ↗

2026-10-03 1 article

2026-09-16 · 💻Trypanosomal CLK1, 14-3-3/ERα, SARS-CoV-2 Mpro (screening library design) · Journal of Chemical Information and Modeling (ACS)

Scope library: a chemical space between fragments and full molecules for tractable hits

Targeted protein degradation Hit finding & screening SAR & lead optimizationCheminformaticsInfectious disease Protein–protein interactions

Scope defines a chemical space between fragments and elaborated low-molecular-weight library compounds, emphasizing compound quality, ligand efficiency and selected property ranges to keep hits tractable for medicinal chemistry.
Scope ring systems (ring-X-ring) organize the space and guide intentional functionalization toward low- to mid-micromolar potency; a roughly 10k-compound set suits medium-throughput SPR, native MS, DSF and biochemical screens.
Case studies on trypanosomal CLK1, a 14-3-3/ERα molecular glue project and SARS-CoV-2 Mpro show comparable primary hit rates to diversity sets, with better ligand efficiency after counter-screening.

💡 Points

Honestly, the pitch I like here is hits you can actually start working on: a library sitting between fragments and full-size compounds, built around ligand efficiency, with ring-X-ring cores that tell you where to functionalize next. A 10k set fits SPR, native MS and DSF, so a smaller group can run it. The open question is whether primary hit rates on par with diversity sets, with better LE after counter-screening, hold up beyond three targets, and how much the scaffold constraint narrows novelty.

The Scope Concept: Defined Chemical Space for Identifying Ligand-Efficient, Tractable Hits Applied to Drug Discovery Targets Trypanosomal CLK1, 14–3–3/Estrogen Receptor-α, and SARS-CoV-2 Mpro ↗

2026-10-02 1 article

2026-09-12 · 💻AKR1B10 (co-folding-based virtual screening) · Journal of Chemical Information and Modeling (ACS)

Boltz-2 Co-folding Narrows an 867-Compound Library to 40 Picks, and Most Turn Out Active

Virtual screening & docking AI & machine learning

The authors tested whether a co-folding model can replace docking for triage, applying Boltz-2 to the cancer-linked enzyme AKR1B10.
From a chemically diverse library of 867 candidates, they chose 40 by predicted affinity plus a diversity filter.
About 70% of those tested showed inhibition at low-micromolar or better, and several submicromolar inhibitors with novel scaffolds emerged.

💡 Points

A 70% confirmation rate from 40 compounds is a striking number for virtual screening, though the starting library was already small and the target is an enzyme with a lipophilic, forgiving pocket. I'd want to see the same experiment on a harder target and against a docking baseline run on identical compounds. Still, a tractable assay count and novel chemotypes make this an easy workflow to copy for hit finding.

Efficient Discovery of Potent AKR1B10 Inhibitors Using Co-folding Model Boltz-2 ↗

2026-09-30 1 article

2026-09-24 · 💻Computational binding free energy (MM/GBSA) / PTM-aware design · Journal of Medicinal Chemistry (ACS)

PLAC Dataset Tests Whether MM/GBSA Can Predict How PTMs Shift Ligand Binding

Virtual screening & docking Molecular simulation

The authors curated PLAC, an experimentally validated set of ligand activity changes caused by post-translational modifications; affecting PTMs tend to sit closer to the ligand than neutral ones.
They benchmarked end-point MM/GBSA protocols across MD lengths and internal dielectric constants.
Longer MD with a higher dielectric gave the best correlation with experiment, while a lower dielectric classified enhance/weaken/neutral better.

💡 Points

Nobody builds SAR around a PTM state, yet it can quietly explain why a potent binder fails in cells. A curated benchmark is worth more to me than yet another scoring tweak, because we finally have something to test against. The dielectric trade-off between ranking and classification is a useful warning: pick the protocol for the question you are actually asking.

Predicting the Post-translational Modification Effects on Protein−Ligand Interactions via End-Point Binding Free Energy Calculation: Database Creation and Strategy Optimization ↗

Computational & AI archive →