将想法转化为成果的 AI 协同科学家苗头化合物

All-atom 3D structure prediction of biomolecular complexes using NVIDIA NIM-packaged OpenFold3.

Biomolecular Emulator (BioEmu) — a generative deep-learning model that samples from the approximated equilibrium distribution of 3D structures for a protein monomer given its amino acid sequence.

Target Sequence-Conditioned Generation of Peptide Binders via Masked Language Modeling

HMMER is a C toolkit for biological sequence analysis using profile hidden Markov models (profile HMMs). It is widely used for protein and DNA homology search and is the engine behind Pfam, InterPro, and many large-scale annotation pipelines. The package is a collection of command-line programs (phmmer, jackhmmer, hmmsearch, hmmscan, nhmmer, nhmmscan, hmmbuild, hmmalign, hmmemit, plus utilities).

Multi-modal foundation model for biomolecular structure prediction of proteins, ligands, DNA, RNA, and complexes.

PROPKA predicts the pKa values of ionizable groups in proteins (v3.0) and protein-ligand complexes (v3.1+) based on the 3D structure using an empirical/heuristic method. It also computes folding free-energy and protein charge profiles as functions of pH.

SaProt is a structure-aware protein language model that combines amino-acid tokens with FoldSeek 3Di structural tokens for improved protein representation, zero-shot mutation effect prediction, embedding extraction, and inverse folding.

Structure-based de novo antibody and nanobody design pipeline combining an antibody-finetuned RFdiffusion for backbone design, ProteinMPNN for CDR sequence design, and an antibody-finetuned RoseTTAFold2 for in silico filtering.

MDTraj is a Python library for reading, writing, and analyzing molecular dynamics (MD) trajectories with fast, vectorized routines for RMSD, secondary structure, hydrogen bonds, distances, dihedrals, SASA, radius of gyration and other observables.

HERGAI is a structure-based AI tool for predicting human Ether-a-go-go-Related Gene (hERG) potassium-channel inhibitors. It trains four binary classifiers (RF_BC, XGB_BC, DNN_BC and the stacking ensemble DNN_SC) on PLEC (Protein-Ligand Extended Connectivity) fingerprints extracted from ClassyPose-selected docking poses of small molecules against a hERG receptor structure. DNN_SC is reported as the best-performing model in the paper.

Pocket2Mol: Efficient Molecular Sampling Based on 3D Protein Pockets. Uses equivariant graph neural networks to autoregressively generate 3D ligand molecules conditioned on a protein binding pocket.

MMseqs2 (Many-against-Many sequence searching) is an ultra-fast, sensitive sequence search and clustering suite for protein and nucleotide sequences.