Accelerating Serendipitous Discovery
New process definition to supersede human limitations
Last updated
New process definition to supersede human limitations
Last updated
Data efficiency plays include using less data to train algorithms, differentiating clean data from garble, and algorithms that interpolate behavior through synthetic data.
To mitigate model drift and longtail data challenges, Bee portfolio company builds dynamic asset databases (CGI enhanced digital twins based upon 3d scans), and can produce 10K+ variants of pixel-perfect labeled images overnight.
Trends like Siri for the biology lab, automated pipetting, labs-on-a-chip, high throughput experimentation, virtual CROs, CV, AI, and ML, are streamlining critical but tedious benchwork and generating rich datasets. Using this data to efficiently traverse experimentation helps cut down discovery time to generate breakthroughs consistently.
Bee portfolio company is powering industry-wide scale up via continuous fermentation at bench-scale which informs their ML algorithm, enabling high-throughput (10x faster and cheaper) bio-production.
Data privacy is a customer and enterprise problem. Infrastructure that allows data to be shared reliably and enables ML infrastructure to update without compromising details of provenance is an exciting area of exploration.
ML companies in the healthcare domain (e.g. ) rely on federated learning to get access to more diverse datasets to speed up drug discovery.
In W2, startups typically aspire to achieve just 1 reproducible pilot-scale fermentation run across 2 years, with a special emphasis on downstream processing (DSP - recovering the desired output, e.g. protein). DSP is custom across outputs and therefore the most difficult to scale.
Bee portfolio company is producing cow cheese without the cow by leveraging a proprietary production method of the milk protein, Casein Micelles. The company raised a Series A before launching product in market, an industry norm.