On September 2, 2025 BostonGene, a leader in AI-powered solutions for drug discovery and development, reported the publication of a collaborative study*, "AI-driven multimodal algorithm predicts immunotherapy and targeted therapy outcomes in clear cell renal cell carcinoma," in Cell Reports Medicine (Press release, BostonGene, SEP 2, 2025, View Source [SID1234655677]). The study introduced the largest harmonized transcriptomic and clinical dataset in kidney cancer and presents a model of how advanced multimodal AI solutions can identify previously unknown determinants of response to combinations of immunotherapy and VEGF inhibitors in solid tumors.
Schedule your 30 min Free 1stOncology Demo!
Discover why more than 1,500 members use 1stOncology™ to excel in:
Early/Late Stage Pipeline Development - Target Scouting - Clinical Biomarkers - Indication Selection & Expansion - BD&L Contacts - Conference Reports - Combinatorial Drug Settings - Companion Diagnostics - Drug Repositioning - First-in-class Analysis - Competitive Analysis - Deals & Licensing
Schedule Your 30 min Free Demo!
While immune checkpoint inhibitors (ICIs) and VEGF inhibitor specific tyrosine kinase inhibitors (TKIs) have transformed care, treatment for metastatic clear cell renal cell carcinoma (ccRCC) remains challenged by variable responses, frequent progression and therapy-related toxicities and a lack of guidance on when to apply combination versus single agent therapy. Previous predictive models showed promise, but lacked reproducibility and biological interpretability across independent cohorts.
Leveraging more than 3,600 patient samples and harmonized clinical data, researchers developed a multimodal foundation model trained on extensive real-world datasets that integrates genomics, transcriptomics and tumor microenvironment (TME) profiling. The digital twin-like model generated representations of patient biology. It uncovered five novel Harmonized Immune Tumor Microenvironment (HiTME) subtypes — distinct categories of ccRCC defined by unique immune infiltration patterns, genomic alterations and prognostic outcomes. Importantly, these subtypes were validated with spatial proteomics, ensuring predictions mapped directly to tumor biology rather than functioning as a "black box."
Using the HiTME subtypes, researchers generated clinically interpretable responder scores that correlated with survival outcomes for both ICIs and TKIs across independent cohorts. This led to a decision-tree tool that stratified patients into ICI-preferred, TKI-preferred or non-responder categories, establishing a data-driven methodology to stratify patients by underlying immune and tumor biology, enabling new insights for drug development and clinical research. The framework also revealed a therapy-resistant subgroup characterized by immune-desert phenotypes and angiogenic signaling–highlighting urgent unmet needs and new therapeutic avenues.
"This study demonstrates how multimodal foundation models can reshape oncology," said Nathan Fowler MD, Chief Medical Officer at BostonGene "By grounding predictions in tumor biology rather than opaque algorithms, we enable clinicians to understand why patients may or may not respond to treatment — making AI both clinically actionable and scientifically trustworthy."