Weekly BioML Digest [June 22, 2026] (updates!)

Machine Learning × Computational Biology paper compilation

Hey! It's your weekly digest of machine learning papers in CompBio and Drug Discovery. I've made some changes to the digest for improved readability, hope it helps!

Updates:

• 📝 Every paper now includes a short summary to help you quickly assess relevance.
• 🔬 PubMed has been added as a source for peer-reviewed papers, expanding coverage.
• 👥 Long author lists are now truncated in the middle for improved readability.
• 🏛️ Author affiliations are displayed whenever available.

Feedback? Email me at biomldigest@gmail.com.

📚 Peer-Reviewed Journals (Top 20)

1161 matched filters -> 20 selected after LLM relevance + novelty ranking.

• 📡 Deep learning–guided engineering of SpuFz1 and rational miniaturization of ωRNA enables efficient genome editing
  Chen, Shuanghong, Hsiao, Shenlin, ..., Wu, Yuxuan, Liao, Jiaoyang — Nature Communications, 2026-06-18
  Combines deep learning–guided protein optimization with rational ωRNA miniaturization to engineer a compact enFanzor genome-editing system achieving up to 81.9% editing in mammalian cells.

• 📰 Automated High-Throughput Virtual Screening of Catalysts via Templated Organic Reaction Pathway Construction: A Case Study on Suzuki-Miyaura Coupling Reaction.
  Zi-Xing Guo, Jin-Peng Tang, Zhen-Xiong Wang, Qi-Ming Liang, ..., Cheng Shang, Sheng-Ye Zhang, Lin Chen, Zhi-Pan Liu — Journal of the American Chemical Society, 2026-06-19
  Introduces a self-learning diffusion + neural PES framework to auto-generate reaction pathways and screen 6,883 Pd-phosphine catalysts for Suzuki–Miyaura coupling at ~$0.01 per catalyst.
  Affiliations: State Key Laboratory of Porous Materials for Separation and Conversion, Collaborative Innovation Center of Chemistry for Energy Material; State Key Laboratory of Organometallic Chemistry, Shanghai Institute of Organic Chemistry

• 📰 Public antibody clonotypes and deep learning identify SARS-CoV-2 and HIV broadly neutralizing antibodies in immune repertoires.
  Lizhi Zhou, Zhili Yu, Shutian Lin, Yanan Jiang, ..., Hai Yu, Ying Gu, Shaowei Li, Ningshao Xia — Cell reports, 2026-06-19
  ClonoDeep integrates public clonotypes with a deep sequence model to mine repertoires for broadly neutralizing antibodies, discovering bnAbs for SARS-CoV-2 and HIV without antigen panning.
  Affiliations: State Key Laboratory of Vaccines for Infectious Diseases, Xiang An Biomedicine Laboratory; National Institute of Diagnostics and Vaccine Development in Infectious Diseases, National Innovation Platform for Industry-Education Integration in Vaccine Research; ...

• 🏛️ Linear-time prediction of proteome-scale microbial protein interactions.
  Andre Cornman, Matt Tranzillo, Nicolo G Zulaybar, Imane Bouzit, Yunha Hwang — Proceedings of the National Academy of Sciences of the United States of America, 2026-06-23
  FlashPPI aligns microbial interactors in a residue-aware latent space from metagenomic language models, enabling linear-time proteome-wide PPI screening with structure-level accuracy at minutes scale.
  Affiliations: Tatta Bio, Cambridge; Microbiology Graduate Program, Massachusetts Institute of Technology; ...

• 📰 Learning High-Resolution Protein Embeddings from Multimodal Data via Self-Supervised Integration.
  Yong-Jia Liang, Qian-Yi Wang, Qian Zhou, Yingying Xu — Journal of chemical information and modeling, 2026-06-18
  self-SSGI learns high-resolution protein embeddings by self-supervised multimodal integration of sequence, structure, GO, and images, surpassing state-of-the-art on multiple protein annotation tasks.

• 📰 PAIRMAP: A Unified Geometry-Aware Pairwise-Map Framework for Molecular Representation Learning.
  Zhejiong Wang, Zhengjun Hu, Lichen Zhu, Yifei Wu, Yanhong Chen, Yinghui Jiang, Haotong Sun, Ying Yu — Journal of chemical information and modeling, 2026-06-18
  Demonstrates learnable pairwise geometric encodings and constrained triangular attention outperform large pretrained models for molecular property prediction across domains.

• 📰 Fitness translocation: improving variant effect prediction with biologically-grounded data augmentation.
  Adrien Mialland, Shuzo Fukunaga, Riku Katsuki, Yunfei Dong, Hideki Yamaguchi, Yutaka Saito — Bioinformatics (Oxford, England), 2026-06-20
  Introduces fitness translocation, a biologically grounded data augmentation using PLM embedding offsets from homologs to boost variant-effect prediction under scarce data.
  Affiliations: Artificial Intelligence Research Center, National Institute of Advanced Industrial Science and Technology (AIST); University of Toronto.; ...

• 📰 Protein-Nucleic Acid Binding Site Prediction Using Interpretable Kolmogorov-Arnold Networks with Hypergraph Representation Learning.
  Yangfeng Zhu, Guicong Sun, Weimin Zhu, Yongxian Fan, Zeheng Wu, Xianchen Zheng, Xiaoyong Pan — Bioinformatics (Oxford, England), 2026-06-20
  IKANbind fuses pLMs, hypergraph neural networks, and interpretable Kolmogorov–Arnold Networks to predict protein–nucleic acid binding residues with mechanistic feature attribution.
  Affiliations: School of Computer Science and Information Security, Guilin University of Electronic Technology; Institute of Image Processing and Pattern Recognition, Shanghai Jiao Tong University

• 📰 Benchmarking MS/MS Featurization Strategies for Machine Learning-Driven Metabolite Structure Annotation.
  Roger Giné, Iván Pérez-López, Josep M. Badia, Jordi Capellades, Oscar Yanes — Journal of the American Society for Mass Spectrometry, 2026-06-16
  A broad benchmark finds adaptive binning and DreaMS embeddings best support deep models for MS/MS-based structure annotation, stressing ppm accuracy for reliable retrieval.

• 📰 Machine learning framework for GW calculations across molecular dynamics trajectories
  R. Abdelghany, Chih-En Hsu, H. Hsueh, Yuan-Hong Tsai, Ming-Chiang Chung — Machine Learning: Science and Technology, 2026-06-17
  Demonstrates that pretrained NIPs can be “hacked” as fixed feature extractors for property models, outperforming end-to-end DNNs under data scarcity.

• 📰 Surface-Enhanced Raman Spectroscopy: A Game Changer for Metabolomics Research.
  Xinyuan Bi, Xing Yi Ling, Jian Ye — Nano letters, 2026-06-17
  Positions SERS as a metabolomics platform, highlighting AI-assisted targeted/untargeted phenotyping and remaining gaps to achieve “true” metabolomic breadth.
  Affiliations: School of Biomedical Engineering, Shanghai Jiao Tong University; School of Chemistry, Chemical Engineering and Biotechnology; ...

• 📡 A universal deep learning framework for empowering nanopore identification by reinforcing temporal signals
  Li, Ming, Li, Minmin, ..., Ning, Hanwen, Qing, Guangyan — Nature Communications, 2026-06-18
  SEDA-Former enhances nanopore molecular identification via multi-scale temporal enhancement and adaptive attention, achieving robust classification across structurally similar analytes.

• 🚀 Discovery of TYR inhibitors from de novo molecular generation to dual-track lead optimization: “Competition” between AI and chemists
  Yinyan Sun, Jiahui Wang, Wenchao Chen, Xiaoying Jiang, ..., Xiaotian Niu, Bin Ju, Jianan Guo, Renren Bai — Science Advances, 2026-06-19
  An RL-based generative pipeline discovers TYR inhibitors and compares expert-guided vs AI-driven optimization, yielding potent, nonintuitive chemotypes with validated antipigmentation activity.

• 📰 Carbonyl-Modulated Lowest Unoccupied Molecular Orbital Energy Directs Machine Learning-Assisted Screening of Electrolyte Additives Toward Ultra-Stable Zinc Metal Anodes.
  Le Zhang, Shuyu Bi, Xijun Liu, Qiangchao Sun, Xionggang Lu, Hongwei Cheng — Advanced materials (Deerfield Beach, Fla.), 2026-06-19
  GNN + SHAP identifies carbonyl electron localization as the key driver in additive screening, selecting α-ketoglutaric acid to form gradient SEI and extend Zn anode longevity.
  Affiliations: School of Materials Science and Engineering & State Key Laboratory of Advanced Refractories, Shanghai University

• 📰 Understanding the precipitation mechanism in pentavalent vanadium electrolytes through deep learning potential molecular dynamics.
  Chenkai Mu, Chenbo Zhan, Tianyu Li, Xianfeng Li — Chemical science, 2026-06-17
  Deep potential MD reveals V(V) precipitation in vanadium flow batteries proceeds via SN2-like hydroxyl dehydration; anion coordination strategy suppresses precipitation, validated experimentally.
  Affiliations: Division of Energy Storage, Dalian National Laboratory for Clean Energy; School of Chemistry and Chemical Engineering, University of Chinese Academy of Sciences Beijing 100049 China.; ...

• 📰 Condensate Growth Analysis Platform for Proteins Using Ultra-Widefield Dark-Field Microscopy and Image Analysis.
  Kiyoto Kamagata, Ren Fujita, Eriko Mano, Nanami Hirashita, Ryu-Suke Nozawa — The journal of physical chemistry. B, 2026-06-18
  Ultra-widefield dark-field microscopy with image analysis tracks ~10k condensates, quantifying fusion-dominated growth in FUS and HP1α systems without fluorescent labels.
  Affiliations: Faculty of Engineering, Gifu University; Graduate School of Natural Science and Technology, Gifu University; ...

• 📰 PAIRMAP: A Unified Geometry-Aware Pairwise-Map Framework for Molecular Representation Learning.
  Zhejiong Wang, Zhengjun Hu, Lichen Zhu, Yifei Wu, Yanhong Chen, Yinghui Jiang, Haotong Sun, Ying Yu — Journal of chemical information and modeling, 2026-06-18
  Demonstrates learnable pairwise geometric encodings and constrained triangular attention outperform large pretrained models for molecular property prediction across domains.

• 📰 Construction of an Interpretable Regression Model for Yield Prediction and Mechanistic Insight Enabled by Automated Reaction Path Exploration.
  Takahiro Doba, Yu Harabuchi, Yuuya Nagata, Satoshi Maeda — Journal of the American Chemical Society, 2026-06-19
  Uses AFIR-generated intermediate energy descriptors to build linear, interpretable yield predictors linking mechanistic intermediates to reaction outcomes.
  Affiliations: International Research Center for Elements Science, Institute for Chemical Research; Institute for Chemical Reaction Design and Discovery (WPI-ICReDD), Hokkaido University; ...

• 📰 Machine Learning-Assisted Surface Ligand Engineering Strategy for Enhanced Sensitivity of Immunoassay Platform.
  Jinbo Cao, Tiemei Li, Yao Wang, Hengheng Xiong, ..., Qingqing Deng, Chongwen Wang, Xiaogang Hu, Li Wang — Analytical chemistry, 2026-06-16
  Ligand charge-transfer engineering plus ML decoding yields a trimetallic nanozyme with HRP-surpassing activity and DL-enabled, ultrasensitive P. aeruginosa immunoassays.
  Affiliations: Guangzhou Key Laboratory of Analytical Chemistry for Biomedicine, GDMPA Key Laboratory for Process Control and Quality Evaluation of Chiral Pharmaceuticals; College of Food and Bioengineering, Henan International Joint Laboratory of Food Green Processing and Quality Safety Control; ...

• 📰 Deep-Learning-Enabled SEM Image Segmentation Coupled with Laser Confocal Raman Microscopy: Elucidating Microstructure and Drug Spatial Distribution in Leuprorelin Acetate Microspheres
  Wei Zhang, Zhihong Xu, Li Jiang, Xiaohu Tang, Chao Wang, aiping wang, Wanhui Liu — Pharmaceuticals, 2026-06-18
  DL-based SEM segmentation coupled with 3D Raman microscopy correlates microstructure and API distributions to controlled-release behavior in PLGA microspheres.

🧬 Preprints (arXiv + bioRxiv)

44 matched filters -> 20 selected after LLM relevance + novelty ranking.

• 🧬 RareFold: Structure prediction and design of proteins with noncanonical amino acids
  Li, Q.; Daumiller, D.; Zuo, F.; Marcotte, H.; Pan-Hammarstrom, Q.; Bryant, P. — bioRxiv, 2026-06-16
  Introduces RareFold, a deep learning architecture for structure prediction and design across 20 canonical and 29 noncanonical amino acids, enabling unified modeling of diverse chemistries. Demonstrates de novo peptide binders with ncAAs validated by HDX-MS and low-micromolar affinities.
  Affiliations: Stockholm University

• 🧬 Robust Conditional Diffusion with Noisy Templates for Antibody Sequence-Structure Design
  Liu, p.; Zhang, J.; Yan, C. — bioRxiv, 2026-06-18
  Presents NT-ABDiff, a reliability-aware conditional diffusion model that jointly designs antibody CDR sequences and structures while robustly handling sparse or corrupted CDR templates. Improves CDR-H3 recovery and RMSD under template shifts for antibody design against novel antigens.
  Affiliations: Nanjing University

• 📄 Contextualizing Biological Language Models across Modalities via Logit-Space Contrastive Alignment
  Yanjun Shao, Yundi Chen, Yashvi Patel, Aurelien Pelissier, María Rodríguez Martínez — arXiv, 2026-06-17
  LOGICA performs contrastive alignment directly in output-logit space to context-condition biological language models while preserving token likelihood interfaces. Boosts performance on protein–ligand binding, TCR–peptide activity, and drug-conditioned resistance prediction and supports interpretable, mutation-local scoring.

• 📄 Be Your Own Teacher: Steering Protein Language Models via Unsupervised Reward Optimization
  Lanqing Li, Shentong Mo, Yang Yu, Pheng-Ann Heng — arXiv, 2026-06-17
  Introduces unsupervised reward optimization to steer protein language models without labels via task-agnostic proxy rewards combining intrinsic uncertainty and semantic consistency. Achieves controllable protein generation that outperforms DPO/KTO and approaches oracle performance on OOD prompts.

• 📄 Agentic Discovery of Non-Canonical Antimicrobial Peptides with AMPGAN v3
  Jay Jung, Xiaohan Zhang, Shenghan Song, Mahmoud Sayedahmed, ..., Severin T. Schneebeli, Matthew J. Wargo, Jianing Li, Safwan Wshah — arXiv, 2026-06-15
  AMPGAN v3 is a multi-objective conditional GAN that generates AMPs with D-amino acids and terminal modifications, stabilized via dual discriminators. Integrated into an agentic discovery workflow, it yields experimentally validated peptides with Gram-positive activity.

• 🧬 Regulatory network hubs guide dynamic human lineage specification
  Takeuchi, C.; Sivakumar, S.; Sundarrajan, A.; Wang, Y.; ...; Chahrour, M. H.; Kraus, W. L.; Munshi, N. V.; Hon, G. C. — bioRxiv, 2026-06-15
  Combines large-scale Perturb-seq with a transformer to predict transcription factor perturbations driving altered regulatory networks from patient transcriptomes. Reveals dynamic hub-centric regulatory architectures guiding human lineage specification and enables disease mechanism inference.
  Affiliations: UT Southwestern Medical Center

• 📄 Toward Controllable Catalyst Inverse Design via Large-Scale Autoregressive Pretraining
  Dong Hyeon Mok, Jonggeol Na, Seoin Back — arXiv, 2026-06-16
  Develops a property-conditioned autoregressive GPT pretrained on 133M catalyst structures to generate heterogeneous catalysts with controlled adsorbate types and binding energies. Achieves high structural validity and improves target binding-energy matching, accelerating reaction-specific catalyst discovery.

• 📄 MassSpecGym in the Wild: Uncovering and Correcting Evaluation Pitfalls in AI-Driven Molecule Discovery
  Hongxuan Liu, Roman Bushuiev, Ivy Lightheart, Mrunali Manjrekar, ..., Runzhong Wang, David Healey, Tomáš Pluskal, Connor W. Coley — arXiv, 2026-06-17
  Audits evaluation pitfalls in MS/MS-driven molecule discovery and releases MassSpecGym v1.5 to fix data leakage, shortcut learning, and metric issues. Establishes more trustworthy benchmarking for spectra-to-molecule ML methods.

• 📄 Calibrating Generative Models to Feature Distributions with MMD Finetuning
  Nathaniel L. Diamant, Brian L. Trippe — arXiv, 2026-06-17
  Introduces kCGM, an MMD-based fine-tuning with KL regularization to calibrate generative models to target feature distributions. Improves feature matching while maintaining validity on antibiotics and extends to protein and DNA generation with only feature-level supervision.

• 📄 ASTEROID: A Spatiotemporal Information Transformer for Forecasting Multi-Step Time Series of Molecular Dynamics
  Kexin Wu, Luonan Chen, Renxiao Wang — arXiv, 2026-06-16
  ASTEROID is a spatiotemporal Transformer that forecasts multi-step MD atomic coordinates by modeling multiscale spatial and temporal dependencies. Outperforms existing predictors and reduces the computational burden of QM-derived molecular dynamics.

• 📄 How Post-Training Shapes Biological Reasoning Models
  Lukas Fesser, Hanlin Zhang, Michelle M. Li, Eric Wang, Bryan Perozzi, Shekoofeh Azizi, Sham M. Kakade, Marinka Zitnik — arXiv, 2026-06-15
  Dissects how continued pretraining, supervised fine-tuning, and RL reshape generalization in biological reasoning models across genomics, transcriptomics, and proteins. Finds RL on strong SFT checkpoints with aligned rewards can recover OOD performance, guiding post-training design for bio LMs.

• 📄 TxBench-PP: Analyzing AI Agent Performance on Small-Molecule Preclinical Pharmacology
  Hannah Le, Ramesh Ramasamy, Alex Urrutia, Mahsa Yazdani, Tim Proctor, Kenny Workman — arXiv, 2026-06-17
  TxBench-PP is a verifiable benchmark where AI agents must analyze real preclinical pharmacology assay files and return structured, auto-graded conclusions. Reveals current agents fail to robustly recover program decisions, setting a realistic target for agentic systems in drug discovery.

• 🧬 Antibody-Antigen Affinity Prediction with Chain-Aware Protein Language Modeling
  Singh, H.; Malhotra, A.; Srivastava, S. P.; SINGH, R. K.; Gorantla, R. — bioRxiv, 2026-06-21
  AbAffinity is a chain-aware, three-stream sequence model that fuses frozen ESM-2 embeddings with gated cross-attention to predict antibody–antigen affinity without structures. Outperforms recent sequence-based baselines and provides residue-level attributions for paratope/epitope signals.
  Affiliations: Shiv Nadar Institution of Eminence (Deemed to be University), Delhi-NCR

• 🧬 FeatureMSEA: Metabolic Feature-based Metabolite Set Enrichment Analysis
  Liu, Y.; Wang, Y.; Huan, T.; Shen, X. — bioRxiv, 2026-06-19
  FeatureMSEA introduces feature-rank metabolite set enrichment that integrates multi-evidence annotations, permutation tests, and iterative refinement to work directly from untargeted LC–MS features. Extracts pathway-level signals without requiring confident metabolite IDs and reduces annotation ambiguity.
  Affiliations: Nanyang Technological University

• 🧬 immgenT: A Comprehensive Reference of Convergent T-cell States in the Mouse
  Magill, I.; Casey, O.; Mallah, D.; Panigrahi, S. S.; ...; Huh, J. R.; Iliev, I.; Jabri, B.; Jameson, — bioRxiv, 2026-06-17
  Uses a deep generative model on joint RNA/protein single-cell data to build a reusable mouse T-cell atlas spanning 8 lineages and 107 states. Provides tools to map external datasets consistently, standardizing T-cell state interpretation across contexts.
  Affiliations: University of Chicago

• 🧬 Buffered sweat microfluidics with AI-enabled translation for clinically actionable blood urea estimation and renal risk stratification
  Zhang, J.; Shen, Z.; Xu, M.; Ge, Y.; ...; Li, N.; Ma, P.; Peng, Z.; Zhao, Y. — bioRxiv, 2026-06-18
  Combines a buffered sweat microfluidic patch with a physiology-informed, AI-enabled calibration model to translate sweat urea into blood urea estimates. Achieves high concordance with gold-standard labs and enables renal risk stratification from noninvasive measurements.
  Affiliations: 4. Dyson School of Design Engineering, Imperial College London

• 🧬 Predicting Mouse Lifespan-Extending Chemical Compounds with Machine Learning
  Belikov, A. V.; Ribeiro, C.; Farmer, C. K.; Petrascheck, M.; de Magalhaes, J. P.; Freitas, A. A. — bioRxiv, 2026-06-17
  Trains Random Forest ensembles on drug-target annotations to predict mouse lifespan-extending compounds and screens DrugBank to prioritize candidates. Experimentally validates six compounds that extend C. elegans lifespan, highlighting GPCR and hormonal signaling axes.
  Affiliations: Genomics of Ageing and Rejuvenation Lab, Institute of Inflammation and Ageing

• 🧬 Biological meaning in protein embedding space is resolution-dependent
  Zong, L.; Ren, J.; Li, Y.; Finn, R. D.; Wang, J. — bioRxiv, 2026-06-15
  Shows that biological meaning in protein embedding neighborhoods is resolution-dependent, with different optimization targets preserving distinct hierarchy-level semantics. Provides guidance for interpreting and aligning PLM spaces to enzyme and family-level biology.
  Affiliations: European Bioinformatics Institute; University of Bath

• 🧬 Large datasets and machine learning models fail to capture extremophile enzyme melting and optimum temperatures
  Gault, S. — bioRxiv, 2026-06-15
  Demonstrates that ML predictors of enzyme melting and optimum temperatures systematically fail for extremophiles due to biased and error-laden training data. Calls for curated, proteome-scale extremophile measurements to enable reliable thermostability modeling.
  Affiliations: University of Edinburgh

• 🧬 SMLMFlow: Improving Structural Resolution in Single Molecule Localization Microscopy with Flow Matching
  Bauer, S.; Panconi, L.; Cunha, I.; Latron, E.; Sage, D.; Peters, R.; Griffie, J. — bioRxiv, 2026-06-15
  SMLMFlow unites a graph neural network and hierarchical Transformer with flow matching to deblur SMLM point clouds. Improves structural resolution and downstream quantification of cellular filaments and nano-clusters.
  Affiliations: Stockholm University, Science of Life Laboratory