Arc Virtual Cell Challenge: Modeling Gene Silencing with Arc's STATE

AI Impact Summary

The Arc Virtual Cell Challenge presents a complex modeling task: predicting the effect of gene silencing on cell types using a sparse RNA sequencing dataset. Participants will leverage a dataset of ~300k single-cell profiles, including ~38k unperturbed control cells, and a model based on a transformer architecture (ST) and a BERT-like autoencoder (SE) to simulate gene silencing effects. The challenge requires understanding of concepts like CRISPR, alternative splicing, and technical noise, alongside the use of models like Llama and ESM2, representing a significant technical hurdle for engineers without a strong biology background.