Journal: bioRxiv

Article Title: Phenotype-Guided In Silico Molecular Generation Using Large Language Models

doi: 10.64898/2026.01.03.697483

Figure Lengend Snippet: a , Schematic illustration of fibroblast–myofibroblast state transitions and the application of GEMGen to the fibrosis transcriptional program. Fibroblast activation is associated with upregulation of fibrosis markers including ACTA2 , FN1 , and COL1A1 . The fibrosis-associated transcriptional signature is used as input to GEMGen to generate candidate small molecules predicted to attenuate fibroblast activation. b , Chemical structures of representative GEMGen-generated compounds. Upper rows: GEMGen-recovered previously reported molecules. Lower rows: previously unreported molecules selected for experimental validation (HG-9-91-01 and indirubin-3′-monoxime) c , qPCR analysis of FN1 transcript levels in TGFβ-stimulated LX-2 cells treated with HG-9-91-01 (left) and indirubin-3′-monoxime (right), normalized to DMSO control levels; SB431542 (10 µM) is included as a positive control. IC50 values are indicated. d , Immunofluorescence analysis of fibronectin deposition in fibroblasts treated as indicated. Nuclei are stained with DAPI (blue), and fibronectin is shown in yellow.

Article Snippet: Cells were then treated with 10 μM SB431542, 200 nM HG-9-91-01, or 3 μM indirubin-3′-monoxime (TargetMol) in DMEM containing 0.1% FBS for 12 hours followed by stimulation with 10 ng/mL TGF-β for 48 hours.

Techniques: Activation Assay, Generated, Biomarker Discovery, Control, Positive Control, Immunofluorescence, Staining