#31 · Genomics
Interpreting cancer variants via imaging, pathogenicity, and single-cell expression
The dominant thread is computational interpretation of tumor genomic and transcriptomic data, with learned models used to relate variants, copy number, expression programs, and phenotypes. A foundation-model approach correlates Evo 2 variant severity scores with TCGA tumor radiomics, reporting 46 new renal cancer genes. A benchmark of 49 missense pathogenicity predictors on 768 cancer genes finds oncogenic gain-of-function variants under-scored and supplies a per-gene calibration map. scOPE transfers frozen, driver-associated expression axes from bulk tumors to single-cell transcriptomes with a ground-truth-free confidence score. CNV-Finder and ALFAssay address detection in genotyping array and cfDNA data, respectively. What is new is the use of sequence-model variant scores as a bridge to imaging phenotypes, and per-gene rather than global calibration of pathogenicity predictors.
Published papers
CNV-Finder: streamlining copy number variation discovery.
CNV-Finder, an LSTM-based deep learning pipeline with human-in-the-loop refinement, detects copy number variants from genotyping array LRR/BAF signals, cutting…
ALFAssay: A feed‑forward neural network for quantitative fragmentomics‑based ctDNA profiling in breast cancer.
ALFAssay, a feed-forward neural network estimating ctDNA fraction in breast cancer from cfDNA fragmentation features in shallow WGS, validated against ichorCNA…
Preprints
Foundation-model-guided radiogenomic discovery linking cancer genomes to cancer scans
Evo 2 variant severity scores correlated with TCGA tumor radiomics to discover gene-imaging associations, including 46 new renal cancer genes.
An openly licensed benchmark and per-gene calibration map for missense pathogenicity predictors on activating cancer drivers
Openly licensed benchmark of 49 missense pathogenicity predictors on 768 cancer genes shows oncogenic GOF variants are under-scored; provides per-gene…
scOPE identifies which driver-associated expression programs transfer from bulk tumors to single cells
scOPE transfers frozen, driver-associated expression axes learned from bulk tumors onto single-cell transcriptomes, with a ground-truth-free confidence score…
Topic history
| Date | Rank |
|---|---|
| #31 | |
| #35 | |
| #22 | |
| #10 | |
| #11 | |
| #12 | |
| #20 | |
| #29 | |
| #8 | |
| #7 | |
| #2 | |
| #7 | |
| #1 | |
| #1 | |
| #6 | |
| #18 | |
| #6 | |
| #32 | |
| #12 | |
| #4 | |
| #15 | |
| #6 | |
| #16 | |
| #22 | |
| #21 | |
| #23 | |
| #12 | |
| #11 | |
| #20 | |
| #13 | |
| #18 | |
| #1 | |
| #6 | |
| #16 | |
| #9 | |
| #17 | |
| #39 |