Stanford and EMBL release SMFNet, a diffusion model for single-molecule chromatin accessibility

Stanford's Anshul Kundaje lab, in collaboration with Arnaud Krebs' group at EMBL Heidelberg, released the SMFNet preprint on bioRxiv on November 6. SMFNet is the first method to use a diffusion model to predict single-molecule chromatin accessibility from DNA sequences, enabling "one sequence, multiple chromatin states" predictions, with an official open-source implementation released alongside.

Confirmed

• SMFNet is a seq2func discrete diffusion model that takes a DNA sequence (2048bp) as input and generates the distribution of single-molecule footprint (SMF) reads over the central 1024bp region, with per-read, per-position annotation
• The preprint was a collaboration between the EMBL Heidelberg Krebs group and the Stanford Kundaje group; co-first authors include Nate (Nathaniel) Diamant and Till, and the preprint is titled "SMFNet reveals principles of combinatorial t…"
• The model decodes single-molecule rules of transcription factor assembly, predicts combinatorial TF assembly, and was validated across 15 perturbation experiments
• The official open-source implementation is available on GitHub under the MIT license, accompanied by an explainer video from the authors

Why it matters

• Traditional chromatin accessibility prediction stays at the population-average level; SMFNet pushes prediction to single-molecule resolution, capturing the multiple chromatin states a single sequence can produce
• By combining generative modeling with perturbation-based validation, this work provides a new tool for studying the sequence-encoded rules of combinatorial transcription factor regulation, and showcases the potential of diffusion models for genomic function prediction tasks

2026-11-06 ~ 2026-11-06 · 7 related posts

• SMFNet: a diffusion model predicting single-molecule chromatin accessibility from DNA — anshulkundaje · 2026-11-06
• SMFNet Preprint: Discrete Diffusion Model Decodes Combinatorial TF Assembly, Validated Across 15 Perturbations — anshulkundaje · 2026-11-06
• [source] Stanford Lab Open-Sources SMFNet, a Discrete Diffusion Model for Single-Molecule Footprinting — anshulkundaje · 2026-11-06
• SMFNet and ChAI-Net: sequence models for single-molecule data probe transcription factor assembly — anshulkundaje · 2026-11-06
• [source] SMFNet: a diffusion model fits single-molecule chromatin state distributions at 2Kb context — anshulkundaje · 2026-11-06

2 near-duplicate retellings: anshulkundaje · anshulkundaje