Chang Ho Ahn

Although genetic sequencing is routine in cancer care, translating a tumor’s complex mutation profile into actionable treatment decisions remains a central challenge. MutationProjector is pretrained on a large corpus of genomic alterations across more than 30,000 tumors, integrated with extensive molecular knowledge. The resulting projection reveals a tumor’s altered molecular pathways, facilitating model interpretation, and it accurately reconstructs held-out mutations, demonstrating model generalization. When applied to predict immunotherapy or chemotherapy resistance across multiple cancer types and cohorts, MutationProjector achieves or exceeds state-of-the-art performance in all contexts. It identifies unexpected biomarkers, including KMT2D mutation in immunotherapy sensitivity and joint alteration of SMARCA4 and STK11 in immunotherapy resistance. These results establish a unifying framework for connecting tumor genotypes to biological mechanisms and therapeutic outcomes. Significance:This work describes MutationProjector, a tumor genomic foundation model pretrained on more than 30,000 tumors. It maps mutations to biological states for diagnosis and treatment, achieving state-of-the-art performance in predicting therapy resistance and identifying biomarkers such as KMT2D (sensitivity) and SMARCA4/STK11 coalterations (resistance).

Figure S1 shows mutational heterogeneity of the pre-training dataset. Figure S2 shows detailed model architecture. Figure S3 shows inferring the status of tumor infiltrating lymphocytes. Figure S4 shows the analysis of latent representations. Figure S5 shows the prediction of drug response using clinical biomarkers. Figure S6 shows the prediction of metastasis-related tasks. Figure S7 shows the method for computing feature importance. Figure S8 shows the analysis of high model attention.

Table S1 contains the similarities of interaction pairs between networks. Table S2 contains the predictive performance in pretraining tasks. Table S3 contains the feature importance scores for anti-cancer drug response tasks. Table S4 contains the gene alteration frequencies in predicted responders and non-responders in immunotherapy treated lung cancer patients. Table S5 contains gene co-alteration frequencies. Table S6 contains number of patients with tumor infiltrating lymphocyte status information. Table S7 contains the prediction performance of anti-cancer drug response tasks.

Aims: To characterize the heterogeneity of pathological responses in the breast and regional lymph nodes after neoadjuvant pembrolizumab‐based chemoimmunotherapy in early‐stage triple‐negative breast cancer (TNBC) and to explore their clinical implications. Methods and results: A total of 203 patients with TNBC treated with neoadjuvant chemotherapy combined with pembrolizumab followed by surgery were included. Histopathological features of baseline biopsy and post‐treatment specimens, including residual carcinoma patterns and tumour‐bed features, were systematically reviewed and correlated with treatment response and clinical outcomes. Pathological complete response (pCR) was achieved in 63.1% of patients. Higher histological grade, increased tumour‐infiltrating lymphocytes (TILs), and higher Ki‐67 labelling index were independently associated with pCR. Residual carcinoma patterns and tumour‐bed changes were heterogeneous, with residual disease most commonly presenting as a solitary, confined focus (51.4%). Fibrotic tumour beds were more frequently observed in non‐pCR cases, whereas stromal elastosis was enriched among pCR cases. Nodal responses were also heterogeneous, including complete nodal response (23.6%), non‐response (8.4%), and mixed nodal response (7.4%). Disease recurrence was more frequent among patients without pCR, and among those with multifocal baseline tumours, higher cT stage, rare histological subtypes, low TILs and residual nodal disease. Conclusions: This real‐world study provides a comprehensive pathological characterization of breast and nodal responses after neoadjuvant chemoimmunotherapy in early TNBC, highlighting heterogeneity beyond pCR and ypN staging and offering a more detailed characterization of treatment‐response patterns.

While genetic sequencing is routine in cancer care, translating a tumor’s complex mutation profile into actionable treatment decisions remains a central challenge. MutationProjector is pre-trained from a large corpus of genomic alterations across 30,000+ tumors, integrated with extensive molecular knowledge. The resulting projection reveals a tumor’s altered molecular pathways, facilitating model interpretation, and it accurately reconstructs held-out mutations, demonstrating model generalization. When applied to predict immunotherapy or chemotherapy resistance across multiple cancer types and cohorts, MutationProjector achieves or exceeds state-of-the-art performance in all contexts. It identifies unexpected biomarkers, including KMT2D mutation in immunotherapy sensitivity and joint alteration of SMARCA4 and STK11 in immunotherapy resistance. These results establish a unifying framework for connecting tumor genotypes to biological mechanisms and therapeutic outcomes.

Olanzapine is a second-generation antipsychotic drug associated with the development of type 2 diabetes. However, the development and remission of diabetes with glutamic acid decarboxylase (GAD) antibody (Ab) positivity related to olanzapine treatment have not been reported. Here, we present the case of a 33-year-old man taking olanzapine who was diagnosed with diabetes with GAD Ab positivity complicated by diabetic ketoacidosis (DKA). After discontinuation of the drug and tight glycemic control, his diabetes remitted, and the GAD Ab became negative spontaneously after 18 months. The classification of diabetes is challenging because of overlapping autoimmune and metabolic features. Olanzapine may have contributed to pancreatic β-cell apoptosis, which could potentially trigger autoimmunity and the development of type 1 diabetes-like features (DKA as the initial presentation, a low C-peptide level, and GAD Ab positivity). The rapid amelioration of hyperglycemia and weight reduction may be responsible for the remission of diabetes after the discontinuation of olanzapine.

Baseline characteristics of the test sets (test set A, B, and C).

Figure S6. Prediction scores for EGFR mutation prediction between two magnifications in test set C (n = 2,261).

Performance of the ensemble model to predict EGFR mutation status across test set A (n = 1,461) and test set B (n = 599). ROC curves for test set A (A–D) and test set B (E–I) are shown. For each test set, the following subgroup analyses were performed: entire dataset (A and E), specimen type (surgical resection and biopsy; B and F), EGFR mutation subtypes (E19del, L858R point mutations, E20ins, T790M, and other mutations; C and G) and scanner type (D and H). For test set B, data source subgroups are additionally presented (I). The sample size (“n”) for each subgroup reflects the number of cases and controls. For EGFR subtypes, each ROC curve was generated using cases positive for the specific EGFR subtype (with or without co-mutations) as EGFR mutant and all EGFR wild-type cases as EGFR wild type. The data represent the AUC together with corresponding 95% CIs.

Baseline characteristics of the training set and tuning set.