Mola is a sequence-to-distribution model of single-molecule chromatin accessibility. Given 10 kb of DNA, it generates the population of chromatin fibers that sequence encodes, one molecule at a time. The web server samples molecules for any 10 kb region or sequence variant and shows them with FiberHMM nucleosome and footprint calls.

Angler is a compact splicing model that matches state-of-the-art accuracy while being much faster, enabling real-time scoring and exhaustive, genome-wide saturation mutagenesis. It predicts donors and acceptors jointly, scoring each site in the context of its neighbours, and so captures exon definition and the competition among closely-spaced sites. The interactive web server provides splicing predictions in real time, allowing users to explore the effects of sequence alterations.

Seq Browser is a genome browser built around real-time sequence-model inference in the browser: small models run locally on the DNA in view, and their outputs draw as ordinary tracks, re-scored as you pan, zoom, flip strand, or edit the sequence. It ships with Puffin and MotifMatch.

Orca-Fly extends the Orca framework to Drosophila, predicting multiscale 3D genome organization directly from genomic sequence. It enables sequence-based prediction and analysis of 3D genome architecture in the fly.

TriFlow models the structure-conditioned sequence landscape for large-scale protein design, generating diverse sequences consistent with a target backbone structure. Its architecture combines global attention with discrete flow matching for fast, few-step sequence generation, enabling flexible and scalable protein design directly from backbone structure.

Across science we routinely observe only noisy or indirect measurements while the true distribution of interest remains latent. Given a known noise or observation model, Inverse Flow recovers that latent distribution directly from the observations, without any ground-truth samples. Its two algorithms, Inverse Flow Matching and Inverse Consistency Model, extend diffusion and consistency models to arbitrary forward processes and complex noise models.

Puffin is an interpretation-focused sequence model for transcription initiation in the human genome which is also applicable to other mammalian species. Puffin can predict basepair-resolution transcription initiation signals using only genomic sequence as input, and more importantly, analyze the sequence basis of any transcription start site (TSS) at motif and basepair levels.

Puffin-D is a prediction-focused deep learning sequence model for basepair-resolution transcription initiation signals. Puffin-D takes 100kb input and accurately predicts the transcriptional signal strength. The prediction-focused deep learning model Puffin-D is available at puffind.zhoulab.io.

For most use cases, we recommend the interactive Puffin web server at puffin.zhoulab.io. The prediction-focused deep learning model Puffin-D is available at puffind.zhoulab.io.

Dirichlet Diffusion Score Model (DDSM) is a continuous-time diffusion framework designed specifically for modeling discrete data such as biological sequences. We introduce a diffusion process defined in probability simplex space with stationary distribution being the Dirichlet distribution. This makes diffusion in continuous space natural for modeling discrete data. We demonstrated applications to Sudoku solving and promoter sequence design.

Orca is a deep learning sequence modeling framework for multiscale genome interaction prediction from kilobase to beyond whole-chromosome scale. Orca allows predicting genome structural impacts of any genomic variants, including very large structural variants, and designing virtual genetic screens to probe the sequence basis of genome 3D organization. A version of Orca for leukemia-related cell lines is also available in the webserver, along with a GitHub repository for scoring the regulatory impact of structural variants across leukemia subtypes.

Sei is a framework for systematically predicting sequence regulatory activities and applying sequence information to human genetics data. Sei provides a global map from any sequence to regulatory activities, as represented by 40 sequence classes, and each sequence class integrates predictions for 21,907 chromatin profiles (transcription factor, histone marks, and chromatin accessibility profiles across a wide range of cell types). Sei has been used for interpreting tissue-specific regulatory signals from GWAS and predicting individual pathogenic variant effects.

GraphDR is a method for single-cell data visualization and representation; StructDR is a method for generalized trajectory inference (StructDR). The unique features of these methods are 1. linear interpretability, which is similar to PCA and eases comparisons across datasets, 2. allowing statistical inference of confidence sets for the trajectories, and 3. unifying the inference of clusters (0-dimensional), trajectories (1-dimensional), and surfaces (2-dimensional) structures. GraphDR is also scalable to more than 10 million cells.

ExPecto is a framework for ab initio sequence-based prediction of mutation gene expression effects and disease risks. Precomputed variant effects can be browsed and downloaded here. You can access deep learning sequence model-based variant chromatin effect prediction from the webserver here (thanks to the Flatiron Institute Genomics team!).

Selene is a Python library and command line interface for developing deep neural networks for biological sequence data. Our aim for Selene is to accelerate development and application of deep learning sequence models in biology. The development of Selene is led by Kathy Chen.

For analyzing impact of regulatory mutations to disease, we developed deep learning sequence models for molecular-level effects at chromatin level and RNA-binding protein level and Disease Impact Scores summarizing molecular level effects. Precomputed ASD mutation effects can be browsed and downloaded here.

DeepSEA is a deep learning-based algorithmic framework for predicting the chromatin effects of sequence alterations with single nucleotide sensitivity. DeepSEA can accurately predict the epigenetic state of a sequence, including transcription factors binding, DNase I sensitivities, and histone marks in multiple cell types. It can be further utilized to predict the chromatin effects of sequence variants and prioritize regulatory variants.

We developed a machine learning method to provide genome-wide, quantitative spatiotemporal gene expression predictions. The new method we developed is structured in silico nano-dissection, a lineage-aware probabilistic graphical model that predicts gene expression in >200 tissue-developmental stages. FIND (Fly in silico nanodissection) is a webserver for exploring the gene expression predictions.

Accelerating Systematic Prediction of Variant Effects and Sequence Interpretation with Multiplexer Models. Multiplexer is a framework for deep learning sequence model-based prediction and interpretation.

A set of lightweight, browser-based utilities for sequence motif analysis. Paste a DNA or RNA sequence to scan it for transcription-factor and RNA-binding motifs with FIMO-style exact p-values, or search a motif against JASPAR, HOCOMOCO, CIS-BP, Vierstra, and CIS-BP-RNA using cross-correlation and Tomtom comparison. Everything runs client-side — no installation and nothing is uploaded — and each tool also downloads as a single self-contained HTML file that works offline.