@scihan AI yüzünden büyük şirketler daha senior ararken küçük şirketler - startuplar daha junior ve üniversite mezunu gençleri işe alma eğiliminde . Ama çevremde gördüğüm genel bir saturasyon var dünyada/türkiyede genel beyaz yaka işlerinde
@DeryaTR_ bioarxiv also do not take submissions made by a student , regardless of their submission quality, I see this is some kind of a gatekeeping. Also arxiv won’t take any submissions without an endorsement that is also gatekeeping :/
Current DTA models memorize dataset-specific patterns instead of learning what drives binding. Top-performing models drop from 0.910 AUROC in-distribution to 0.5 cross-dataset essentially random.
Anchor Transfer Learning fixes this by reformulating DTA as a comparison @ebetica
On Davis kinases with verified zero drug overlap: CI 0.642, AUROC 0.669 while pairwise baselines collapse.
74% of predictive variance is orthogonal to protein similarity the model learns drug-specific binding patterns, not just nearest-neighbor matching.