scTab: Scaling cross-tissue single-cell annotation models.

Identifying cellular identities is a key use case in single-cell transcriptomics. While machine learning has been leveraged to automate cell annotation predictions for some time, there has been little progress in scaling neural networks to large data sets and in constructing models that generalize well across diverse tissues. Here, we propose scTab, an automated cell type prediction model specific to tabular data, and train it using a novel data augmentation scheme across a large corpus of single-cell RNA-seq observations (22.2 million cells). In this context, we show that cross-tissue annotation requires nonlinear models and that the performance of scTab scales both in terms of training dataset size and model size. Additionally, we show that the proposed data augmentation schema improves model generalization. In summary, we introduce a de novo cell type prediction model for single-cell RNA-seq data that can be trained across a large-scale collection of curated datasets and demonstrate the benefits of using deep learning methods in this paradigm.

Utility of public knowledge bases for the interpretation of comprehensive tumor molecular profiling results.

With the growing use of comprehensive tumor molecular profiling (CTMP), the therapeutic landscape of cancer is rapidly evolving. NGS produces large amounts of genomic data requiring complex analysis and subsequent interpretation. We sought to determine the utility of publicly available knowledge bases (KB) for the interpretation of the cancer mutational profile in clinical practice. Analysis was performed across patients who previously underwent CTMP. Independent interpretation of the CTMP was performed manually, and then, the recommendations were compared to ones present in KBs (OncoKB, CIViC, CGI, CGA, VICC, MolecularMatch). A total of 222 CTMP reports from 222 patients with 932 genomic alterations (GA) were identified. For 368 targetable GA identified in 171 (77%) of the patients, 1381 therapy recommendations were compiled. Except for CGA, therapy ESCAT LOE I, II, IIIA and IIIB therapy options were equally represented in the majority of KB. Personalized treatment options with ESCAT LOE I-II were provided for 35 patients (16%); MolecularMatch/CIViC allowed to collect ESCAT I-II treatment options for 34 of them (97%), OncoKB/CGI-for 33 of them (94%). Employing VICC and CGA 6 (17%) and 20 (57%) of patients were left without ESCAT I or II treatment options. For 88 patients with ESCAT level III-B therapy recommendations: only 2 (2%), 3 (3%), 4 (5%) and 6 (7%) of patients were left without options with CIViC, MolecularMatch, CGI and OncoKB, and with VICC-12 (14%). Highest overlap ratio was observed for IIIA (0.81) biomarkers, with the comparable results for LOE I-II. Meanwhile, overlap ratio for ESCAT LOE IV was 0.22. Public KBs provide substantial information on ESCAT-I/R1 biomarkers, but the information on ESCAT II-IV and resistance biomarkers is underrepresented. Manual curation should be considered the gold standard for the CTMP interpretation.