This Week In Cheminformatics: Issue #021

conformal prediction for molecular retrieval from MS2, how expensive should your conformer generation workflows be and a long list of papers

Highlights

---

Conformer Generation Workflows for COSMO-RS Calculations: Are They All the Same?

Gomes et al. systematically evaluate whether expensive conformer generation is strictly necessary for COSMO-RS thermodynamic predictions. The authors tested four pipelines. They used RDKit ETKDG/MMFF94 to compute-heavy DFT and semiempirical GOAT workflows across 926 experimental solubility data points. Surprisingly, for general high-throughput screening of typical drug-like molecules, the RDKit-Direct (ETKDGv3), as the authors call it, performed comparably to the complex pipelines, with mean absolute errors around 0.90 log units. Authors note that for highly flexible molecules with more than five rotatable bonds or systems prone to solvent-dependent intramolecular hydrogen bonding, choice of method dictates the result drastically.

Reliable Molecular Retrieval from Mass Spectra Using Conformal Prediction

This paper by Rakhshaninejad et al. applies conformal prediction to candidate-based molecular retrieval from LC-MS/MS data. Usually, retrieval pipelines are evaluated using metrics like top-k accuracy, which summarize performance at the dataset level but fail to provide a spectrum-specific reliability statement. To address this, the authors construct prediction sets that contain the true molecule with a user-specified probability, effectively quantifying uncertainty for individual spectra. They evaluated marginal and conditional conformal prediction across in-distribution, partially shifted, and fully out-of-distribution scenarios on the MassSpecGym benchmark. As this framework operates directly on the output scores of a retrieval model without requiring any architectural modifications, it is a good read for anyone looking to implement uncertainty quantification in their annotation pipelines.

Long List

---

Cheminformatics

• Algorithm-driven, phenotype-directed bioactive molecular discovery

• strainedSMILES2xyz: a workflow for reliable 3D structures of strained molecules from SMILES

• Breaking glycolysis: allosteric hotspots for multi-target drug repurposing

• Terminal-Flip Monte Carlo: Accelerating the Convergence of Molecular Dynamics and Alchemical Free Energy Calculations

• AdBand‐CTNet: A Hybrid Convolution Transformer Framework for Precise NIR‐Based Determination of 1,8‐Cineole in Large Cardamom

• kNNPC (k‐Nearest Neighbor Algorithm Per‐Class): Classification Method Predicting Extrapolation Regions With Reasonable Probability

• Deep learning for assay nuisance compound detection using a gated co-attention graph embedding model (CAGE-Fusion)

• QUASAR: A Universal Autonomous System for Atomistic Simulation and a Benchmark of Its Capabilities

• MEGCAM: MEta-Graph and Causal Attention Method for Drug Repurposing on Heterogeneous Drug–Target–Disease Knowledge Network

• RxnBench: A Multimodal Benchmark for Evaluating Large Language Models on Chemical Reaction Understanding from the Scientific Literature

• PL-PatchSurfer3: improved structure-based virtual screening for structure variation using 3D Zernike descriptors

• Integrated computational modeling of chalcone-based inhibitors targeting carbonic anhydrase I and II: 3D-QSAR, molecular docking, and dynamics simulations

• Molspectra: a general framework for multi-spectra prediction from molecular structures

• Mutation-Driven Remodeling of the LRRK2 Kinase Free-Energy Landscape and Its Consequences for Conformational Transitions and Inhibitor Binding Affinity

• Mechanistic Insights into Cystic Fibrosis Transmembrane Conductance Regulator Function from Molecular Dynamics Analysis of Electrostatic Interactions

• Improving Protein Structure Prediction Using Integrative Cryo-EM and Ion Mobility Mass Spectrometry Modeling

• Dynamics of Electronically Inelastic and Reactive Collisions of O(1D) with N2 Based on Machine-Learned Coupled Potential Energy Surfaces

• Machine-Learning Predictions of Photoluminescence in Molecules Exhibiting Thermally Activated Delayed Fluorescence with Implicit Experimental Validation

• MolSculptor: An Adaptive Diffusion–Evolution Framework Enabling Generative Drug Design for Multitarget Affinity and Selectivity

• Molecular Dynamics Workflows to Compute Large-Scale Sets of Absolute Binding Free Energies Aiding Drug Candidate and Binding Pose Selection

• Policy-Based Active Learning for Efficient Molecular Identification

• A Spectral Matching Algorithm Based on the Wasserstein Metric

• Bridging predictive reliability and explainability: a multi-representation deep learning framework for chemical space analysis of immune bioassays

• A systematic evaluation of protein allosteric site prediction tools with independent datasets

• ULCYP: A Multitask Model for Predicting P450 Inducers Based on Positive-Unlabeled Learning

• Evaluating Molecular Docking Programs for RNA-Targeted Ligand Screening: Influence of Binding Modes and Ligand Types

• ConfDTI: Structure Confidence-Guided Multimodal Fusion for Drug–Target Interaction Prediction

• Comprehensive annotation and analysis of human microproteins by human microprotein atlas platform

• Functional Groups Are All You Need for Chemically Interpretable Molecular Property Prediction

• Assessment of Alphafold Protein Models for Small-Molecule Ligand Docking versus Co-Folding

• CAHNetF-DTP: A Community-Aware Heterogeneous Network-Based Embedding Framework for Drug-Target Interaction Prediction

• Cryo-Electron Microscopy Structural Ensemble Optimization Using Individual Particles

• Novel Molecules Generation Using Graph Generative Adversarial Networks

• An SE(3)-equivariant and dynamic multi-modal engine advancing from PTM site prediction to network understanding

MedChem

• Discovery of a High-Potency Balanced GLP-1/GIP Dual Agonist by Molecular Dynamics Evolution

• Discovery of the First Highly Potent, Selective, Irreversible Small-Molecule Factor XIIa Inhibitor for Treating Sepsis

• De novo design of anticancer 4-thiazolidinone derivatives: a generative framework shaped by activity cliffs

• Decoding Prolonged Residence Time of 5-HT2A Receptor Antagonists: Insights from Ritanserin Derivatives

Other

• Host–Guest Complexation through Geometric Self-Optimization in 12Cycloparaphenylene

• pi-stack optimizer: framework for the design of one-dimensional supramolecular systems

• A diagnostic analysis of B3LYP thermochemical errors: physical origins, system dependence, and practical implications

Palate Cleanser

---

• The Story of the Woodward–Hoffmann Rules — Very cool !!!

Luca Dellanna@DellAnnaLuca

The reactions of many researchers on finally being held responsible for having read the very paper they submitted are... something.

12:13 PM · May 16, 2026 · 299K Views

---

271 Replies · 887 Reposts · 8.81K Likes

François Fleuret@francoisfleuret

Awesome. Seriously, people are harsh with this platform, but if you are careful with whom you follow, it is a constant stream of awesomness.

Eric Jang @ericjang11

For the last few months I've been working on a from-scratch implementation of AlphaGo, a 2016 AI breakthrough that inspired me to get into deep learning. My casual understanding of AlphaGo was "search-augmented deep neural networks trained with self-play", but I wanted to go

9:41 AM · May 16, 2026 · 138K Views

---

13 Replies · 45 Reposts · 1.17K Likes

Jamie Birch@birch_js

Yeah good luck searching for "c" in your codebase a month from now

Jamon @jamonholmgren

Unpopular opinion: 1- or 2-letter variable names in focused, obvious contexts are totally fine.

12:44 AM · May 15, 2026 · 616K Views

---

311 Replies · 294 Reposts · 11.6K Likes

LaurieWired@lauriewired

30.9% of genetics papers data are kind of trash because of Excel’s aggressive auto-formatting. Until 2023, there was no global option to disable data conversion. For example, the human SEPT family (1-14) of genes is directly related to cell division and cancer research. I’ll

5:20 PM · May 13, 2026 · 105K Views

---

106 Replies · 299 Reposts · 2.69K Likes

andrew arruda@andrewarruda

“E-mails 'hurt IQ more than pot'” - CNN, Friday, April 22, 2005

unusual_whales @unusual_whales

"Prolonged AI use may make it harder to think critically and creatively," per the Economist

2:29 PM · May 13, 2026 · 285K Views

---

103 Replies · 500 Reposts · 4.12K Likes

Jack Fields@OrdinaryInds

The AI bros are discovering CSS stylesheets. What a fucking time to be alive.

Nicolas Bustamante @nicbstme

A lot of people are arguing that HTML burns more tokens than markdown. It's true but you can save at least 40% by externalizing the CSS to a template with <link rel="stylesheet" href="./styles.css">. This style.css is your formatting so the LLM will never output CSS again. I

4:48 PM · May 12, 2026 · 348K Views

---

73 Replies · 364 Reposts · 6.03K Likes

Sofia Sanchez@SofiasBio

Chat, they're already working on better animations !!! it's all good now😅

1:32 PM · May 13, 2026 · 813 Views

---

1 Reply · 15 Likes

Imran S. Haque (@ihaque@{bsky,genomic}.social)@ImranSHaque

me whenever i dip my toes back into quantum chemistry

7:18 PM · May 11, 2026 · 367 Views

---

11 Likes

spidey@lochan_twt

“We used to memorize the whole syntax”

4:57 AM · May 11, 2026 · 103K Views

---

53 Replies · 448 Reposts · 6.52K Likes

Nic Barker@nicbarkeragain

One of the biggest problems with using LLMs as a google replacement for programming, is that getting zero relevant results on google used to be a signal that you had the wrong idea about the root cause. Whereas LLMs will happily indulge any terrible idea you suggest.

4:20 AM · May 11, 2026 · 194K Views

---

140 Replies · 621 Reposts · 10.2K Likes

---

---

Best,
Manas