All Competitors

Fine-tuned Enformer derivative that predicts discrete, interpretable cis-regulatory element class annotations (enhancer, promoter, insulator) directly from DNA sequence across human cell types.

Conditions a pretrained DNA sequence embedding on CpG methylation to predict gene regulation across cell types and alleles, generalizing zero-shot to imprinting, X-inactivation, and accessibility.

Transformer framework that models protein-protein interactions at residue resolution, generalizing zero-shot to unseen MHC alleles and sequence-neutral PTMs from one fixed checkpoint.

A 120M-parameter genomic foundation model that learns adaptive DNA token boundaries via H-Net-style dynamic chunking instead of fixed k-mer or byte-pair tokenization.

Self-supervised foundation model that learns reusable representations of cancer genomes from somatic SNVs and copy-number alterations across 33 tumor types.

A zebrafish DNA sequence-to-function model predicting cell-type-specific single-cell expression across 85 cell-type x developmental-timepoint combinations during embryogenesis.

Biohub's 2026 protein language model trained on ~2.8 billion sequences, forming the representation core of its world model of protein biology.

Combines the ProtT5-XL protein language model with protein-specific evolutionary constraints to predict mutational effects on stability, binding, and epistasis—largely zero-shot.

A 4.7B-parameter Mixture-of-Experts genomic foundation model pretrained on ~1.2 trillion nucleotide tokens from human-associated microbial genomes.

A cross-modal transfer-learning model that adapts the ESM-2 650M protein language model to mRNA analysis by swapping amino-acid tokens for codon tokens, applied to mRNA benchmarks without re-training.

Multimodal deep-learning framework that detects and localizes DNA lesions directly from native nanopore sequencing, built on the damage-aware LesionBERT foundation model.

A mechanistic-interpretability framework that trains sparse autoencoders on protein language model embeddings to extract interpretable features for zero-shot variant effect prediction.

A multimodal Q-former that fuses DNA sequence, gene context, protein function, and text into a prefix for a frozen LLM, enabling zero-shot genetic variant interpretation.

A residual VQ-VAE tokenizer that learns a discrete alphabet of protein conformational ensembles from molecular dynamics data, usable as a frozen representation layer for downstream tasks.

A unified bio-language Mixture-of-Experts foundation model spanning DNA, protein sequence and structure, and biological text, applied across eight task families from a single checkpoint.

A pretrained DNA language model combining Mamba state-space layers, gated dilated convolutions, and Fourier-based attention to capture multi-scale genomic regulatory patterns.

Contrastively fine-tuned ESM-2 (35M and 150M) protein language models that produce general-purpose sequence embeddings where biological similarity maps to embedding proximity.

A protein language model that aligns ESM sequence embeddings with molecular-dynamics trajectory embeddings via contrastive learning for zero-shot mutation-effect prediction.

An open autoregressive genomic foundation model (0.5B–8B params) with a 6-mer DNA tokenizer, matching Evo2-7B win rates at far higher throughput.

Generative multimodal foundation model that jointly models DNA, RNA, protein, and cellular context across six biological modalities, with SOTA splicing prediction.

A direction-aware foundation model trained on ~23M bulk RNA-seq differential-expression profiles that simulates coordinated gene dynamics in viral infection.

Long-context (64 kb) RNA foundation model using the Striped-Hyena architecture for zero-shot prediction of transcriptome architecture from unspliced pre-mRNA sequence.

A supervised, chromatin-informed foundation model that predicts regulatory activity directly from plant genomic sequence in Arabidopsis and rice.

A genomic foundation model that learns DNA representations through joint-embedding prediction in latent space rather than nucleotide reconstruction.