OBJECTIVES: Cancer diagnosis comes as a shock to many patients, and many of them feel unprepared to handle the complexity of the life-changing event, understand technicalities of the diagnostic reports, and fully engage with the clinical team regarding the personalized clinical decision-making. MATERIALS AND METHODS: We develop Oncointerpreter.ai an interactive resource to offer personalized summarization of clinical cancer genomic and pathological data, and frame questions or address queries about therapeutic opportunities in near-real time via a graphical interface. It is built on the Mistral-7B and Llama-2 7B large language models trained on a local database trained using a large, curated corpus. RESULTS: We showcase its utility with case studies, where Oncointerpreter.ai extracted key clinical and molecular attributes from deidentified pathology and clinical genomics reports, summarized their contextual significance and answered queries on pertinent treatment options. Oncointerpreter also provided personalized summary of currently active clinical trials that match the patients' disease status, their selection criteria, and geographic locations. Benchmarking and comparative assessment indicated that the model responses were generally consistent, and hallucination, ie, factually incorrect or nonsensical response was rare; treatment- and outcome related queries led to context-aware responses, and response time correlated with verbosity. DISCUSSION: The choice of model and domain-specific training also affected the response quality. CONCLUSION: Oncointerpreter.ai can aid the existing clinical care with interactive, individualized summarization of diagnostics data to promote informed dialogs with the patients with new cancer diagnoses. AVAILABILITY: https://github.com/Siris2314/Oncointerpreter.

MOTIVATION: Cell-type annotation is a time-consuming yet critical first step in the analysis of single-cell RNA-seq data, especially when multiple similar cell subtypes with overlapping marker genes are present. Existing automated annotation methods have a number of limitations, including requiring large reference datasets, high computation time, shallow annotation resolution, and difficulty in identifying cancer cells or their most likely cell of origin. RESULTS: We developed Census, a biologically intuitive and fully automated cell-type identification method for single-cell RNA-seq data that can deeply annotate normal cells in mammalian tissues and identify malignant cells and their likely cell of origin. Motivated by the inherently stratified developmental programs of cellular differentiation, Census infers hierarchical cell-type relationships and uses gradient-boosted \decision trees that capitalize on nodal cell-type relationships to achieve high prediction speed and accuracy. When benchmarked on 44 atlas-scale normal and cancer, human and mouse tissues, Census significantly outperforms state-of-the-art methods across multiple metrics and naturally predicts the cell-of-origin of different cancers. Census is pretrained on the Tabula Sapiens to classify 175 cell-types from 24 organs; however, users can seamlessly train their own models for customized applications. AVAILABILITY AND IMPLEMENTATION: Census is available at Zenodo https://zenodo.org/records/7017103 and on our Github https://github.com/sjdlabgroup/Census.

Nuclei segmentation is a fundamental task in histopathology image analysis. Typically, such segmentation tasks require significant effort to manually generate accurate pixel-wise annotations for fully supervised training. To alleviate such tedious and manual effort, in this paper we propose a novel weakly supervised segmentation framework based on partial points annotation, i.e., only a small portion of nuclei locations in each image are labeled. The framework consists of two learning stages. In the first stage, we design a semi-supervised strategy to learn a detection model from partially labeled nuclei locations. Specifically, an extended Gaussian mask is designed to train an initial model with partially labeled data. Then, self-training with background propagation is proposed to make use of the unlabeled regions to boost nuclei detection and suppress false positives. In the second stage, a segmentation model is trained from the detected nuclei locations in a weakly-supervised fashion. Two types of coarse labels with complementary information are derived from the detected points and are then utilized to train a deep neural network. The fully-connected conditional random field loss is utilized in training to further refine the model without introducing extra computational complexity during inference. The proposed method is extensively evaluated on two nuclei segmentation datasets. The experimental results demonstrate that our method can achieve competitive performance compared to the fully supervised counterpart and the state-of-the-art methods while requiring significantly less annotation effort.

Cancer is a genetic disorder, meaning that a plethora of different mutations, whether somatic or germ line, underlie the etiology of the 'Emperor of Maladies'. Point mutations, chromosomal rearrangements and copy number changes, whether they have occurred spontaneously in predisposed individuals or have been induced by intrinsic or extrinsic (environmental) mutagens, lead to the activation of oncogenes and inactivation of tumor suppressor genes, thereby promoting malignancy. This scenario has now been recognized and experimentally confirmed in a wide range of different contexts. Over the past decade, a surge in available sequencing technologies has allowed the sequencing of whole genomes from liquid malignancies and solid tumors belonging to different types and stages of cancer, giving birth to the new field of cancer genomics. One of the most striking discoveries has been that cancer genomes are highly enriched with mutations of specific kinds. It has been suggested that these mutations can be classified into 'families' based on their mutational signatures. A mutational signature may be regarded as a type of base substitution (e.g. C:G to T:A) within a particular context of neighboring nucleotide sequence (the bases upstream and/or downstream of the mutation). These mutational signatures, supplemented by mutable motifs (a wider mutational context), promise to help us to understand the nature of the mutational processes that operate during tumor evolution because they represent the footprints of interactions between DNA, mutagens and the enzymes of the repair/replication/modification pathways.

MOTIVATION: Somatic variant calling typically requires paired tumor-normal tissue samples. Yet, paired normal tissues are not always available in clinical settings or for archival samples. RESULTS: We present SomVarIUS, a computational method for detecting somatic variants using high throughput sequencing data from unpaired tissue samples. We evaluate the performance of the method using genomic data from synthetic and real tumor samples. SomVarIUS identifies somatic variants in exome-seq data of ∼150 × coverage with at least 67.7% precision and 64.6% recall rates, when compared with paired-tissue somatic variant calls in real tumor samples. We demonstrate the utility of SomVarIUS by identifying somatic mutations in formalin-fixed samples, and tracking clonal dynamics of oncogenic mutations in targeted deep sequencing data from pre- and post-treatment leukemia samples. AVAILABILITY AND IMPLEMENTATION: SomVarIUS is written in Python 2.7 and available at http://www.sjdlab.org/resources/ CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

OBJECTIVE: Currently, there is a disconnect between finding a patient's relevant molecular profile and predicting actionable therapeutics. Here we develop and implement the Integrating Molecular Profiles with Actionable Therapeutics (IMPACT) analysis pipeline, linking variants detected from whole-exome sequencing (WES) to actionable therapeutics. METHODS AND MATERIALS: The IMPACT pipeline contains 4 analytical modules: detecting somatic variants, calling copy number alterations, predicting drugs against deleterious variants, and analyzing tumor heterogeneity. We tested the IMPACT pipeline on whole-exome sequencing data in The Cancer Genome Atlas (TCGA) lung adenocarcinoma samples with known EGFR mutations. We also used IMPACT to analyze melanoma patient tumor samples before treatment, after BRAF-inhibitor treatment, and after BRAF- and MEK-inhibitor treatment. RESULTS: IMPACT Food and Drug Administration (FDA) correctly identified known EGFR mutations in the TCGA lung adenocarcinoma samples. IMPACT linked these EGFR mutations to the appropriate FDA-approved EGFR inhibitors. For the melanoma patient samples, we identified NRAS p.Q61K as an acquired resistance mutation to BRAF-inhibitor treatment. We also identified CDKN2A deletion as a novel acquired resistance mutation to BRAFi/MEKi inhibition. The IMPACT analysis pipeline predicts these somatic variants to actionable therapeutics. We observed the clonal dynamic in the tumor samples after various treatments. We showed that IMPACT not only helped in successful prioritization of clinically relevant variants but also linked these variations to possible targeted therapies. CONCLUSION: IMPACT provides a new bioinformatics strategy to delineate candidate somatic variants and actionable therapies. This approach can be applied to other patient tumor samples to discover effective drug targets for personalized medicine.IMPACT is publicly available at http://tanlab.ucdenver.edu/IMPACT.

Solid tumor samples typically contain multiple distinct clonal populations of cancer cells, and also stromal and immune cell contamination. A majority of the cancer genomics and transcriptomics studies do not explicitly consider genetic heterogeneity and impurity, and draw inferences based on mixed populations of cells. Deconvolution of genomic data from heterogeneous samples provides a powerful tool to address this limitation. We discuss several computational tools, which enable deconvolution of genomic and transcriptomic data from heterogeneous samples. We also performed a systematic comparative assessment of these tools. If properly used, these tools have potentials to complement single-cell genomics and immunoFISH analyses, and provide novel insights into tumor heterogeneity.

Integrative analyses of genomic, epigenomic and transcriptomic features for human and various model organisms have revealed that many such features are nonrandomly distributed in the genome. Significant enrichment (or depletion) of genomic features is anticipated to be biologically important. Detection of genomic regions having enrichment of certain features and estimation of corresponding statistical significance rely on the expected null distribution generated by a permutation model. We discuss different genome-wide permutation approaches, present examples where the permutation strategy affects the null model and show that the confidence in estimating statistical significance of genome-wide enrichment might depend on the choice of the permutation approach. In those cases, where biologically relevant constraints are unclear, it is preferable to examine whether key conclusions are consistent, irrespective of the choice of the randomization strategy.

Mechanisms that generate transcript diversity are of fundamental importance in eukaryotes. Although a large fraction of human protein-coding genes and lincRNAs produce more than one mRNA isoform each, the regulation of this phenomenon is still incompletely understood. Much progress has been made in deciphering the role of sequence-specific features as well as DNA-and RNA-binding proteins in alternative splicing. Recently, however, several experimental studies of individual genes have revealed a direct involvement of epigenetic factors in alternative splicing and transcription initiation. While histone modifications are generally correlated with overall gene expression levels, it remains unclear how histone modification enrichment affects relative isoform abundance. Therefore, we sought to investigate the associations between histone modifications and transcript diversity levels measured by the rates of transcription start-site switching and alternative splicing on a genome-wide scale across protein-coding genes and lincRNAs. We found that the relationship between enrichment levels of epigenetic marks and transcription start-site switching is similar for protein-coding genes and lincRNAs. Furthermore, we found associations between splicing rates and enrichment levels of H2az, H3K4me1, H3K4me2, H3K4me3, H3K9ac, H3K9me3, H3K27ac, H3K27me3, H3K36me3, H3K79me2, and H4K20me, marks traditionally associated with enhancers, transcription initiation, transcriptional repression, and others. These patterns were observed in both normal and cancer cell lines. Additionally, we developed a novel computational method that identified 840 epigenetically regulated candidate genes and predicted transcription start-site switching and alternative exon splicing with up to 92% accuracy based on epigenetic patterning alone. Our results suggest that the epigenetic regulation of transcript isoform diversity may be a relatively common genome-wide phenomenon representing an avenue of deregulation in tumor development.

Abstract Motivation: The biological significance of genomic features is often context dependent. Annotating a particular dataset with existing external data can provide insight into function. Results: We present CruzDB, a fast and intuitive programmatic interface to the University of California, Santa Cruz (UCSC) genome browser that facilitates integrative analyses of diverse local and remotely hosted datasets. We showcase the syntax of CruzDB using microRNA binding sites as examples, and further demonstrate its utility with three biological discoveries. First, DNA replication timing is stratified in gene regions—exons tend to replicate early and introns late during S phase. Second, several non-coding variants associated with cognitive functions map to lincRNA transcripts of relevant function, suggesting potential function of these regulatory RNAs in neuronal diseases. Third, lamina-associated genomic regions are highly enriched in olfaction-related genes, indicating a role of nuclear organization in their regulation. Availability: CruzDB is available at https://github.com/brentp/cruzdb under the MIT open-source license. Contact: [email protected] or [email protected] Supplementary information: Supplementary data are available at Bioinformatics online.