Deepcell and Gilead teamed up to develop a custom AI foundation model for Chinese hamster ovary (CHO) cell line development using morphology data. The partnership leverages Deepcell’s REM-I platform with self-supervised learning to infer cell-line characteristics linked to titer, stability, aggregation, and metabolic profiles—aiming to reduce the time and labor of cell-line development iterations. DeepCyte launched DeeTox Atlas, a single-cell metabolomic reference map designed to capture drug toxicity mechanisms with foundation-model scalability. The company positioned the dataset as a mechanism-based alternative to repeating wet-lab toxicity experiments for every new compound, emphasizing the ability to detect signatures in small cellular subpopulations. Finally, a new pathology-AI reliability framework, TRUECAM, was introduced by researchers at The Hong Kong Polytechnic University, designed to flag when predictions are uncertain so clinicians can take over—addressing trust gaps that can limit deployment even when models appear accurate.