Abstract

Background: Germline mutations contribute to cancer susceptibility, but systematic frameworks to infer cancer type directly from germline profiles remain limited.

Methods: We developed a retrograde prediction framework to identify likely cancer types from germline high-risk variants. High-risk genes were defined as those harboring HIGH-impact or canonical loss-of-function variants. Four complementary strategies were applied: (i) direct intersection with TCGA and cancer stem cell (CSC) gene sets (Case 1), (ii) variant-level scoring (Case 2), (iii) pathway enrichment (Case 3), and (iv) network-based diffusion on protein–protein interactions (Case 4). We tested the framework in five subjects (pt1–pt5), including four cancer patients and one non-cancer individual.

Results: Representative cases highlighted the specificity of each approach: Case 1 (gastric cancer) predicted gastric, breast, and colon cancers; Case 2 (endometrial cancer) predicted breast, colon, and ovarian cancers; Case 3 (triple negative breast cancer) predicted ovarian, colon, and breast cancers; and Case 4 (colon cancer) predicted gastric, colon, and leukemia/lymphoma. The non-cancer subject still yielded gastric, breast, and colon predictions, underscoring both potential false positives and latent susceptibility. The recurrence of gastric, colon, and breast reflects both patient gene distributions and the driver gene richness of these tumor types.

Conclusions: This framework illustrates that germline high-risk variants, when contextualized by curated driver sets, pathways, and networks, can provide early, hypothesis-generating predictions of cancer type. While not diagnostic, this approach may inform risk stratification, surveillance strategies, and future precision prevention efforts.