Tumor Microenvironment

The tumor microenvironment (TME) is a multi-faceted cellular environment that both constrains the evolving tumor and plays a pivotal role in tumor progression and therapeutic response. We are developing computational approaches to profile emergent features of the TME, such as ligand-receptor signaling (Ghoshdastider et al. 2021), metabolism (Rohatgi et al. 2022 ), and tumor purity (Revkov et al. 2023). We are studying biomarkers of drug response and resistance, such as TKI drug resistance in lung cancer (Chua et al. 2021) and response to immunotherapy (Guo et al. 2024).

Cancer Liquid Biopsies

In cancer patients, blood plasma can carry circulating tumor DNA (ctDNA) fragments originating from tumor cells, offering non-invasive access to somatic genetic alterations in tumors. The ctDNA profile of a cancer patient is clinically informative in at least two major ways. Firstly, the profile can provide information about specific actionable mutations that can guide therapy. Secondly, the profile can be used to infer tumor growth dynamics by estimating the amount of ctDNA in the blood. This latter information offers a promising non-invasive approach to track disease progression during clinical trials or therapy, offering a real-time tool to adjust therapy. We are developing machine-learning based approaches to detect and quantify ctDNA in blood samples (Guo et al. 2019, Zhu et al. 2021'>Zhu et al. 2021, Zhu et al. 2025). We are also exploring how ctDNA profiles can be used to study mechanisms of treatment resistance to chemotherapy and targeted therapy.

Integrative Cancer Genomics

We use computational approaches to study cancer biology and precision oncology. We have participated in several large scale tumor characterization projects in The Cancer Genome Atlas (TCGA) (E.g. TCGA ccRCC, Cerami et al. 2012, Gao et al. 2013). We have experience studying tumor whole genomes at scale (e.g. Weinhold, Jacobsen et al. 2014), non-coding mutations (Guo et al. 2018), and non-coding RNAs in cancer (e.g. Jacobsen et al. 2013). We enjoy studying distinct facets of cancer genomics through collaborative projects (e.g. Hakimi et al. 2015, Chen et al. 2016, Montes et al. 2015, Hu et al. 2017, Montes et al. 2020). In Singapore, we are co-investigators in the national lung cancer (e.g. Tan et al. 2017, Chen et al. 2020) and gastrointestinal cancer (Ooi et al. 2020, Xing et al. 2020) consortiums.

cWords - motif discovery

cWords is an algorithm and software suite for analysis of regulatory sequence motifs in gene expression data. Originally developed for analysis of microRNA target sites in 3' UTRs (Jacobsen et al., Genome Research (2010)), but can easily be adapted to other settings. The newest version features an improved and much more parametric approximation (Rasmussen, Jacobsen & Krogh, Silence (2013)).

• URL: version 1.0
• Themes: Web-service, Method, miRNA, Non-coding

miRNA targets in cancer

CancerMiner is a comprehensive resource for discovery and prioritization of functional miRNA-mRNA target interactions in human cancer using molecular datasets from The Cancer Genome Atlas. Candidate interactions can be queried for specific miRNAs or mRNAs, in individual or across multiple cancer types. Our approach uses molecular profiles of >3,000 tumors from 11 human cancer types in The Cancer Genome Atlas. Using this data, we systematically analyzed expression of miRNAs and mRNAs across cancer types to infer recurrent cancer-associated miRNA-target relationships (Jacobsen et al., Nature Structural & Molecular Biology (2013)).

• URL: CancerMiner
• Themes: Web-service, Cancer genomics, miRNA, Non-coding

R + cBioPortal

CGDSR is an R-package to fetch and analyze data from The Cancer Genome Atlas and the cBioPortal for Cancer Genomics. These resources are further described here - Introduction to CGDSR, Cerami et al., Cancer Discovery (2012), Gao et al., Science Signaling (2013).

• URL: CGDSR
• Themes: Web-service, R, Cancer Genomics

Cancer Mutation Calling

Somatic Mutation calling method using a Random Forest (SMuRF) integrates predictions and auxiliary features from multiple somatic mutation callers using a supervised machine learning approach. SMuRF is trained on community-curated matched tumor and normal whole genome sequencing data. SMuRF predicts both SNVs and indels with high accuracy in genome or exome-level sequencing data. Furthermore, the method is robust across multiple tested cancer types and predicts low allele frequency variants with high accuracy. In contrast to existing ensemble-based somatic mutation calling approaches, SMuRF works out-of-the-box and is orders of magnitudes faster. See Huang et al. 2019.

• URL: SMuRF
• Themes: Cancer Genomics, Mutations

MutSpot

Recurrence and clustering of somatic mutations (hotspots) in cancer genomes may indicate positive selection and involvement in tumorigenesis. MutSpot performs genome-wide inference of mutation hotspots in non-coding and regulatory DNA of cancer genomes. MutSpot performs feature selection across hundreds of epigenetic and sequence features followed by estimation of position- and patient-specific background somatic mutation probabilities. MutSpot is user-friendly, works on a standard workstation, and scales to thousands of cancer genomes. Guo et al. 2020.

• URL: MutSpot
• Themes: Cancer Genomics, Mutations

TME Crosstalk

CITE (Crosstalk Interactions within Tumor microEnvironment) provides a database and dynamic visualization of ligand-receptor (LR) signaling interactions within the tumor microenvironment (TME). Crosstalk between ligands and receptors on cancer and stromal cells were estimated in the TME of 20 solid tumor types, based on cancer and stroma specific expression estimated using a tumor transcriptome deconvolution approach. See Ghoshdastider et al. 2020.

• URL: CITE
• Themes: Cancer Genomics, Tumor Microenvironment

TME Metabolic Pathways

The Cancer Metabolic Pathway viewer (CaMP) contains metabolic activity profiles of cancer cells in vivo, non-malignant (stroma) cells within tumors, normal tissues and cancer cells in vitro. Metabolic profiles of cancer cells and stromal cells for 20 solid tumor types is estimated using a data deconvolution approach (Ghoshdastider, 2019) based on transcriptomes of 7865 tumors obtained from the cancer genome atlas (TCGA). Metabolic profiles of normal tissues is estimated using transcriptomes of 815 normal tissue samples from TCGA and genotype-tissue expression (GTEX). 620 cancer cell line from the cancer cell line encyclopedia (CCLE) transcriptomes are used to estimate the metabolic activity of cancer cells in vitro. The database contains KEGG maps of 55 metabolic pathways based on 1845 metabolic genes. See Rohatgi et al. 2020

• URL: CaMP
• Themes: Cancer Genomics, Metabolomics, Tumor Microenvironment

Cancer Genomics Datahub

We have developped the Singapore Oncology Data Portal (OncoSG), building on the cBioPortal from MSKCC. OncoSG enables integration, visualization, analysis, and sharing of cancer genomics datasets from Singapore. We are hosting both a public as well as private (resticted-access) repository.

• URL: OncoSG
• Themes: Cancer Genomics, Database

Circulating Tumor DNA

Profiling of circulating tumor DNA (ctDNA) may offer a non-invasive approach to monitor disease progression. Here, we develop a quantitative method, exploiting local tissue-specific cell-free DNA (cfDNA) degradation patterns, that accurately estimates ctDNA burden independent of genomic aberrations. Nucleosome-dependent cfDNA degradation at promoters and first exon-intron junctions is strongly associated with differential transcriptional activity in tumors and blood. A quantitative model, based on just 6 regulatory regions, could accurately predict ctDNA levels in colorectal cancer patients. Strikingly, a model restricted to blood-specific regulatory regions could predict ctDNA levels across both colorectal and breast cancer patients. Using compact targeted sequencing (<25 kb) of predictive regions, we demonstrate how the approach could enable quantitative low-cost tracking of ctDNA dynamics and disease progression.

• URL: NDRquant
• Themes: Circulating Tumor DNA

Cancer Mutation Calling

Identification of somatic mutations in tumor samples is commonly based on statistical methods in combination with heuristic filters. Here we develop VarNet, an end-to-end deep learning approach for identification of somatic variants from aligned tumor and matched normal DNA reads. VarNet is trained using image representations of 4.6 million high-confidence somatic variants annotated in 356 tumor whole genomes. We benchmark VarNet across a range of publicly available datasets, demonstrating performance often exceeding current state-of-the-art methods. Overall, our results demonstrate how a scalable deep learning approach could augment and potentially supplant human engineered features and heuristic filters in somatic variant calling (Krishnamachari et al. 2022).

• URL: VarNet
• Themes: Cancer Genomics, Mutation Calling

Tumor Purity Estimation

Tumors are complex masses composed of malignant and non-malignant cells. Variation in tumor purity (proportion of cancer cells in a sample) can both confound integrative analysis and enable studies of tumor heterogeneity. Here we developed PUREE, which uses a weakly supervised learning approach to infer tumor purity from a tumor gene expression profile. PUREE was trained on gene expression data and genomic consensus purity estimates from 7864 solid tumor samples. PUREE predicted purity with high accuracy across distinct solid tumor types and generalized to tumor samples from unseen tumor types and cohorts. Gene features of PUREE were further validated using single-cell RNA-seq data from distinct tumor types. In a comprehensive benchmark, PUREE outperformed existing transcriptome-based purity estimation approaches. Overall, PUREE is a highly accurate and versatile method for estimating tumor purity and interrogating tumor heterogeneity from bulk tumor gene expression data, which can complement genomics-based approaches or be used in settings where genomic data is unavailable.

• URL: PUREE
• Themes: Cancer Genomics, Tumor Purity Estimation

Immunotherapy

Immune checkpoint therapy (ICB) has conferred significant and durable clinical benefit to some patients with cancer. However, most patients do not respond to ICB, and reliable biomarkers of ICB response are needed to improve patient stratification. Here, we performed a transcriptome-wide meta-analysis across 1,486 tumors from ICB-treated patients and tumors with expected ICB outcomes based on microsatellite status. Using a robust transcriptome deconvolution approach, we inferred cancer- and stroma-specific gene expression differences and identified cell-type specific features of ICB response across cancer types. Consistent with current knowledge, stromal expression of CXCL9, CXCL13, and IFNG were the top determinants of favorable ICB response. In addition, we identified a group of potential immune-suppressive genes, including FCER1A, associated with poor response to ICB. Strikingly, PD-L1 expression in stromal cells, but not cancer cells, is correlated with ICB response across cancer types. Furthermore, the unbiased transcriptome-wide analysis failed to identify cancer-cell intrinsic expression signatures of ICB response conserved across tumor types, suggesting that cancer cells lack tissue-agnostic transcriptomic features of ICB response.

• URL: IOselect
• Themes: Immunotherapy, TME

Circulating Tumor DNA

Quantification of circulating tumor DNA (ctDNA) levels in blood enables non-invasive surveillance of cancer progression. Fragle is an ultra-fast deep learning-based method for ctDNA quantification directly from cell-free DNA fragment length profiles. We developed Fragle using low-pass whole genome sequence (lpWGS) data from multiple cancer types and healthy control cohorts, demonstrating high accuracy, and improved lower limit of detection in independent cohorts as compared to existing tumor-naïve methods. Uniquely, Fragle is also compatible with targeted sequencing data, exhibiting high accuracy across both research and commercial targeted gene panels. We used this method to study longitudinal plasma samples from colorectal cancer patients, identifying strong concordance of ctDNA dynamics and treatment response. Furthermore, prediction of minimal residual disease in resected lung cancer patients demonstrated significant risk stratification beyond a tumor-naïve gene panel. Overall, Fragle is a versatile, fast, and accurate method for ctDNA quantification with potential for broad clinical utility. Zhu et al. 2025.

• URL: Fragle
• Themes: Circulating Tumor DNA