
Last month I took part in the GI Builder Challenge organised by @GenomeLayer , and it really showed me how much potential there is in efficient DNA sequence design. 🧬 Currently, DNA sequence design is a long process 🕙, with parts often being engineered in isolation—for example, different promoters tested with one fixed CDS—and a lot of trial and error, which can take months. With the rise of sequence-to-function predictors that account for both coding and non-coding sequences, alongside genomic language models, there is an opportunity to streamline biology, reduce trial and error, when designing optimised DNA sequences. In the challenge, I focused only on gene expression, but as more models start predicting different biological modalities, the same idea could be used to design sequences for other functions too, like increased chromatin accessibility. It’s also not limited to mammalian cells; this predictor + generator framework could work across species, opening up applications ranging from agriculture to therapeutics. 🌱💊 Link to the article: substack.com/home/post/p-208… Thank you @julia_kiseleva and @MikhailBurtsev for the opportunity and allowing me to use the GI models. #bioinformatics #AI #computationalbiology #DNA


























