#1 · Tools & infrastructure
Learned models link histology and spatial transcriptomics to predict expression and niches
The dominant thread is learned latent representations that recover structure from spatial transcriptomics and histology. Several items predict spot-level expression from H&E: HESpotEx, a dual-stream graph attention autoencoder, covers up to 5,457 genes, and an EfficientNet-B0 single-patch baseline addresses liver and breast cancer Visium data. STARS applies a Vision Transformer with contrastive learning to reconstruct single-cell expression across Visium and Visium HD, while HAST applies heterogeneity-driven graph learning at subcellular resolution and HESTIA integrates histology with high-resolution data for spatial domain identification. TMEformer and TRINUS extend learned models to in silico perturbation and to separating lineage from niche pressure. Prostate-TriMod adds a tri-modal resource linking virtual H&E, immune and epithelial maps and clinical outcomes. The set is heterogeneous: Bayesian mSGR regression and mass spectrometry imaging manifold learning pursue related goals through different formulations.
Published papers
Evaluation of Physics-Based Pose Prediction and the Role of Receptor Flexibility in a Community Antiviral Blind Challenge.
Physics-based pose prediction using templated molecular growth with hybrid ML/MM potentials; 74% success on SARS-CoV-2 and MERS-CoV main protease, plus…
SMARTFlexDB: a paired apo-holo RNA structure database and analysis platform for RNA conformational flexibility and small-molecule recognition.
SMARTFlexDB is a curated database of 1187 paired apo-holo RNA pocket structures from the PDB, with conformational-change descriptors and a web analysis…
PED in 2027: integrating experimental and predicted protein conformational ensembles at scale.
PED 2027 expands the Protein Ensemble Database to integrate experimental and predicted IDP/IDR conformational ensembles, with APIs, precomputed descriptors,…
DBAASP V4: condition-resolved bioactivity and chemical structures for data-driven peptide and peptidomimetic discovery.
DBAASP v4 is a curated database linking peptides and peptidomimetics to condition-resolved antimicrobial, antibiofilm, anticancer and cytotoxicity data, with…
ARIEL: an openly accessible database of experimental RNA-ligand binding affinities paired with 3D structures.
ARIEL is an open, manually curated database pairing experimental RNA-ligand Kd values with 3D structures, with 278 nonredundant complexes spanning nine orders…
SAbDab2: the structural antibody database in the age of machine learning.
SAbDab2 is an updated public antibody structure database supporting new antibody formats, with ML-grade structures and versioned train/test splits for…
Preprints
Can Folding Models Tell Binders from Bluffers? Evidence from POISK: The Patent-Derived Antibody Dataset
POISK: a patent-derived dataset of 89,074 antibody-antigen records benchmarks six structure predictors on binder vs. decoy discrimination; best model reaches…
DeepPhage: leveraging the power of NGS phage display data, an in-depth analysis for the identification of Eu3+-binding peptides
DeepPhage, an open-source command-line tool, performs comparative statistical analysis and visualisation of phage display NGS data to identify sequence bias…
Topic history
| Date | Rank |
|---|---|
| #1 | |
| #2 | |
| #3 | |
| #5 | |
| #1 | |
| #4 | |
| #5 | |
| #5 | |
| #3 | |
| #3 | |
| #1 | |
| #1 | |
| #1 | |
| #1 | |
| #1 |