Move fast and learn things
Latest publications
A quantitative-genetic decomposition of a neural network
We tested equivalent linear mapping (ELM) on a neural network trained to predict phenotypes from genotypes in simulated data. We show that ELM successfully recapitulates additive and epistatic effects learned by the model, even in data with substantial environmental noise.
Read now
From black box to glass box: Making UMAP interpretable with exact feature contributions
We transform UMAP from a black box into a glass box. By learning the embedding function with a certain type of deep network, we can compute equivalent linear mappings of the input features that exactly reconstruct each embedding, revealing the heretofore hidden logic of UMAP.
Read now
Equivalent linear mappings of deep networks are a promising path for biology
Deep networks make accurate predictions, but their nonlinearity makes them a black box, hiding what they have learned. Here, we look inside the black box and analyze the exact relationships they learn for UMAP embeddings and epistasis in a genotype–phenotype dataset.
Read now
A free, open-access library of high-quality organism illustrations for science communication
We create vector graphics of model organisms and emerging biological research organisms to enhance our publications. We’re sharing these editable graphics under a CC0 license for other scientists to use in figures, slides, teaching materials, or other outputs.
Read now
How do we balance labor and yield in high-throughput protein expression?
We need to test hundreds of in silico protein variant designs in the lab. We have automated purification strategies, but expressing enough protein to purify is low-throughput and rate-limiting. Can we crank up expression without big, unwieldy culture flasks and intensive effort?
Read now
Efficient GFP variant design with a simple neural network ensemble
We designed novel GFPs using a neural network ensemble. We quickly developed an experimental validation procedure in parallel, confirming that several candidates were functional. This rapid loop of in silico generation and lab validation may accelerate protein engineering.
Read now