AI-driven virtual transcriptome screening to identify repurposable combination partners for JIN-A02 in NSCLC.

Abstract

e20528 Background: Combination therapy is a key strategy to overcome drug resistance and improve therapeutic efficacy in non-small cell lung cancer (NSCLC). However, experimentally exploring the full landscape of combination partners for a new therapeutic agent is costly and time-consuming, limiting the pace of rational combination discovery. Transcriptome-guided, scalable prioritization approaches may accelerate identification of clinically actionable combinations and drug repurposing opportunities. Methods: We conducted a large-scale in silico screening to identify potential combination treatment candidates in the context of JIN-A02 treatment. We used RNA-seq data generated from JIN-A02-treated Ba/F3 cells engineered to express EGFR 19del/T790M/C797S to define the JIN-A02-induced transcriptional response signature. Starting from this transcriptomic profile, we virtually generated transcriptomes representing co-inhibition conditions of 15,089 individual targets. These synthetic co-inhibition transcriptomes were systematically compared to assess their ability to enhance or modulate the JIN-A02-associated transcriptional program, including amplification of desired response pathways and suppression of compensatory signaling patterns. Candidate conditions were ranked based on transcriptome-level impact and prioritization of pharmacologically actionable targets. Results: The screening produced a ranked list of putative co-inhibition conditions, and we identified 12 promising combination candidates within the top 100 conditions. Top-ranked candidates showed consistent transcriptomic modulation patterns suggestive of strengthened anti-tumor signaling and attenuation of potential adaptive responses relative to the JIN-A02 signature alone. Notably, one of the top-ranked candidates corresponded to an approved drug, highlighting a drug repurposing opportunity with potential translational advantages. Conclusions: Our results demonstrate the feasibility of virtual transcriptome generation as a scalable framework for discovering rational combination therapies in NSCLC. This strategy enables systematic exploration of a large target space, prioritizes clinically actionable combination hypotheses for JIN-A02, and reveals opportunities for drug repurposing. Follow-up studies will focus on experimental validation of predicted combinations and identification of biomarkers associated with combination sensitivity.