TrendBio

#7 · ML / AI methods

LLM agents ground omics and clinical genomic analysis in executable skills, memory, and…

4 preprints · 2 published · ↑ 2.3× vs 4-wk avg
Items per day, last 8 weekstoday: 6

The dominant thread is coupling LLMs to executable skills, shared laboratory memory, structured query interfaces, or trained predictors so outputs are checkable. ClinGeX composes 105 skills for clinical genomic analysis and reports 71.7% macro-averaged concordance with expert classifications. SMART tests transfer of scRNA-seq cancer findings across six cancer types in 121 studies, while LabWeaver adds laboratory-tier memory and skills on a synovial sarcoma dataset. SPaRTA contributes a 7,500-pair benchmark for text-to-spatial-SQL and analyzes LLM failure modes. SurvXAI layers locally run LLM narratives over Random Survival Forest predictions and SurvSHAP attributions. The Rosalind assessment of GPT-5.2, Claude Sonnet 4.6 and other LLMs against expert and human answers offers a baseline for unaided multi-step reasoning and code writing. The items are not uniform in design: some are agents, others benchmarks or explanation layers.

Published papers

Preprints

Topic history

DateRank
#7
#30
#31
#33

Sponsors