Longevity InTime: Autonomous AI Institute. Anti-Aging Digital Health Immortality Transhumanist AI Channel

AI predicts E. coli colony shape from neighboring IPTG doses On August 27 the authors posted an arXiv preprint, August 27 in which they gave Gemini Co‑Scientist images of genetically altered *E. coli* colonies at several IPTG concentrations, hid the image for one concentration, and asked the model to predict the hidden colony. In the pLac‑rpoS variant IPTG activates the *rpoS* gene that influences swarming; with higher doses the colonies shrink and their radial branches become denser. The control pLac‑gfp strain kept its shape regardless of dose, providing a test of whether the program could recognize stability. Earlier in May Co‑Scientist had suggested genetic factors for testing and helped parse screening results; in this new task it received data from a completed series and was asked to reconstruct the colony shape using only images of the other conditions. The lab grew and scanned the colonies, and the authors sequentially hid images for one concentration at a time. Gemini 3 Pro Image produced 16 colony variants; Gemini 2.5 Pro selected one of them. The physically grown colony at the hidden dose served as an independent check. Predictions were compared to the real colony by average radius, elongation, edge roughness, and roundness. For pLac‑rpoS the first three of the four measurable traits matched the laboratory data, while the generated colonies appeared rounder than the real ones in the roundness metric. The control pLac‑gfp shape remained stable, confirming the program’s ability to detect constancy. Humans defined the task, cultured the bacteria, and captured the images; Co‑Scientist built an interpolation method between known doses. The authors propose applying this workflow in experiments where one biological system is screened across many conditions to decide which measurements to make next. 🔗 Read original →