ArcInstitute/state

State is a machine learning model that predicts cellular perturbation response across diverse contexts

What it solves

State provides a framework for predicting how cells respond to perturbations (such as genetic changes) across different biological contexts. It allows researchers to model cellular state transitions and create high-dimensional embeddings of cellular data to better understand biological responses.

How it works

The project implements two primary model types:

  1. State Transition (ST) Models: These models are trained to predict the outcome of a perturbation. The workflow involves preprocessing raw cellular data (normalizing counts and selecting highly variable genes), training the model using specific dataset configurations, and then performing inference or evaluation on new data.
  2. State Embedding (SE) Models: These models are pretrained to create embeddings of cells. The process involves building a data profile, fitting the embedding model, and transforming new datasets into embedding space. These embeddings can be stored in a vector database (like LanceDB) for similarity queries.

Who it’s for

It is designed for computational biologists and researchers working with single-cell data who need to predict genetic perturbation effects or annotate new datasets using pretrained cellular embeddings.

Highlights

  • Perturbation Prediction: Ability to train models that predict cellular responses to genetic perturbations.
  • Embedding Generation: Tools to pretrain and apply cell embedding models for dataset annotation.
  • Vector Database Integration: Optional support for LanceDB to store and query cellular embeddings.
  • Flexible Configuration: Uses Hydra and TOML files to manage complex dataset splits, including zero-shot and few-shot scenarios.

Related

  • Dispatch
    Arc Virtual Cell Challenge Primer

    The Arc Virtual Cell Challenge tasks participants with training models to predict the effects of gene silencing via CRISPR on cell transcriptomes, utilizing a dataset of 300k single-cell RNA sequencing profiles.

  • Project
    scverse/pertpy

    A framework for analyzing large-scale single-cell perturbation experiments to determine how cells respond to genetic, drug, or environmental stimuli.

  • Project
    huggingface/sentence-transformers

    A framework for computing embeddings, retrieval, and reranking using state-of-the-art dense, sparse, and multi-vector encoder models.

  • Project
    scverse/cellrank

    A modular framework for fate mapping in single-cell data that uses Markov state modeling to estimate cellular dynamics and differentiation probabilities.

  • Dispatch
    Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers

    Hugging Face published a guide on training and finetuning multimodal embedding and reranker models with Sentence Transformers, demonstrating how task-specific finetuning improves performance on retrieval tasks such as Visual Document Retrieval.