3. Biomolecular Property Prediction and Analysis Tools

AlphaGenome: A unified deep learning model from Google DeepMind that takes up to 1 Mb of DNA sequence as input and predicts thousands of functional genomic tracks at up to single-base-pair resolution, including gene expression, transcription initiation, chromatin accessibility, histone modifications, transcription factor binding, chromatin contact maps, and splice site usage/junction strength. Trained on human and mouse genomes, AlphaGenome matches or exceeds the strongest external models in 25 of 26 variant-effect-prediction evaluations, and accurately recapitulates the mechanisms of clinically relevant variants near the TAL1 oncogene.
Paper published: January 29, 2026
Paper | Code (GitHub)

PRIME: A temperature-guided protein language model from Shanghai Jiao Tong University/ShanghaiTech that proposes stabilizing and activity-enhancing mutants for a target protein without requiring prior experimental mutagenesis data on that protein, leveraging implicit thermal-adaptation signal learned across protein families adapted to different environmental temperatures. On a benchmark of 283 deep mutational scanning assays, PRIME outperformed 11 SOTA competing models; in wet-lab validation across five diverse proteins, over 30% of PRIME-recommended mutants outperformed the wild type, including an 8-site LbCas12a variant with +6.5°C melting temperature (retaining trans-cleavage activity) and a 12-site T7 RNA polymerase variant that surpassed the commercial NEB enzyme.
Paper published: November 27, 2024
Paper | Code (GitHub)

EVOLVEpro: A few-shot active learning framework for in silico directed protein evolution, combining a protein language model with a top-layer regression model that learns a protein's activity landscape from as few as 10 experimental data points per round. Benchmarked in silico across 12 protein datasets and applied to five real proteins (a monoclonal COVID antibody, a miniature CRISPR nuclease, a Bxb1 integrase, a prime editor, and T7 RNA polymerase), EVOLVEpro achieved up to 100-fold improvement of desired properties within as few as four rounds of evolution, without requiring structural information or prior data.
Preprint posted: July 18, 2024
Preprint | Code (GitHub)

DDAffinity: A neural network for predicting changes in binding affinity upon multiple mutations.
Paper published: June 28, 2024
Paper | Code (GitHub)

CarbonDesign: A protein sequence design method that adapts the success ingredients of AF2. CarbonDesign utilizes Inverseformer, a network architecture adapted from AlphaFold’s Evoformer. Based on scTM score, sequence recovery rate, and BLOSUM score, CarbonDesign outperformed other published methods, including versions of ProteinMPNN and ESM-IF, on CAMEO and CASP15 test sets, as well as RFdiffusion-generated backbones. CarbonDesign is also capable of predicting the effects of mutations on protein function.
Paper published: May 23, 2024
Paper | Preprint | Code (GitHub) | Code (Code Ocean)

PreMode: A method for gene-specific mode-of-action prediction for missense variants.
Preprint posted: March 17, 2024
Preprint | Code (GitHub)

Evo: A long-context foundation model that generalizes across the central dogma of biology: DNA, RNA, and proteins. Evo is a 7 billion parameter model trained to generate DNA sequences and is capable of prediction and generative tasks, from molecules to whole genomes.
Preprint posted: March 06, 2024
Preprint | Code (GitHub) | Code (PyPI) | Blog | Playground | Colab Notebook

PRESCOTT: A model and webserver for predicting the effect of missense mutations, that rivals AlphaMissense. PRESCOTT is a predictor that combines ESCOTT, an epistatic and structural model of mutational effects, with allele frequencies information from the Genome Aggregation Database (gnomAD). PRESCOTT identifies protein regions most susceptible to mutations, scores individual point mutations, and categorizes mutations as benign, pathogenic, and uncertain. The PRESCOTT model was built on protein structural models, human genomic and exomic data, and huge protein sequences across species.
Preprint posted: Feb 06, 2024
Preprint | Code (GitLab) | Webserver | Docker Image

DeepGlycanSite: A neural network for predicting carbohydrate binding sites on proteins.
Paper published: Jun 17, 2024
Paper | Code (GitHub)

Colabind: A cloud-based method for predicting binding sites using CG MD simulations in the presence of chemical probes.
Paper published: March 21, 2024
Paper | Code (GitHub) | Colab Notebook

GrASP: A method for binding site prediction in protein structures.
Preprint posted: Feb 03, 2024
Preprint | Code (GitHub) | Colab Notebook

DiffPALM: A method for pairing interacting protein sequences based on protein language models trained on MSAs.
Paper published: Jun 24, 2024
Paper | Preprint | Code (GitHub) | Code (Zenodo)

PPIscreenML: A classifier that uses AF2 models to distinguish interacting protein pairs from non-interacting pairs.
Preprint posted: March 17, 2024
Preprint | Code (GitHub)

MAGPIE: An interactive tool for visualization and analysis of interactions between a protein or ligand and its binding partners.
Preprint posted: March 04, 2024
Preprint | Code (GitHub)

Tapioca: A pipeline for de novo prediction of dynamic protein–protein interactions.
Paper published: Feb 15, 2024
Paper | Code (GitHub) | Webserver

AlphaCRV: An AlphaFold-based pipeline for accurate identification of binder topologies by clustering, ranking and visualization of models.
Preprint posted: Feb 08, 2024
Preprint | Code (GitHub)

ARCTIC-3D: An automated tool and webserver to mine and rationalize protein-protein interface information for a given protein. For a given protein, ARCTIC-3D retrieves existing interface information from the PDBe-graph database, and then applies hierarchical clustering on calculated similarity values of interfaces projected on the protein structure. Information generated by ARCTIC-3D can be used as input restraints to docking methods like HADDOCK, and similar to AlphaLink and ColabDock (both using experimental restraints), to drive protein complex modeling with results comparable to prediction by AF-Multimer.
Paper published: Jan 06, 2024
Paper | Code (GitHub) | Webserver

PypKa: A tool and webserver for physics-based and ML-assisted protein pKa estimation and biomolecular structure preparation. PypKa prepares input structures compatible with CHARMM, AMBER and GROMOS ffs. Two additional, alternative methods are provided for pKa estimation, including PypKa, pKAI and pKAI+
Paper published: April 15, 2024
Paper | Code (GitHub) | Webserver

DeepKa: A tool and webserver for protein pKa prediction.
Paper published: March 26, 2024
Paper | Code (GitHub) | Webserver

AlphaGenome: A unified deep learning model from Google DeepMind that takes up to 1 Mb of DNA sequence as input and predicts thousands of functional genomic tracks at up to single-base-pair resolution, including gene expression, transcription initiation, chromatin accessibility, histone modifications, transcription factor binding, chromatin contact maps, and splice site usage/junction strength. Trained on human and mouse genomes, AlphaGenome matches or exceeds the strongest external models in 25 of 26 variant-effect-prediction evaluations, and accurately recapitulates the mechanisms of clinically relevant variants near the TAL1 oncogene.
Paper published: January 29, 2026
Paper | Code (GitHub)

eRMSF: A Python package (built as an MDAnalysis Toolkit) that generalizes root-mean-square-fluctuation (RMSF) analysis beyond time-ordered MD trajectories to any structural ensemble, including AI-generated ones (BioEmu, subsampled AlphaFold2) and experimental ensembles. It supports MDAnalysis-compatible formats, customizable atom/residue/region selections, and a segmentation strategy that computes RMSF over consecutive ensemble segments to produce heatmaps revealing structural heterogeneity that a single averaged RMSF would mask.
Paper published: November 19, 2025
Paper | Code (GitHub)