Ongoing Projects

Analysis of samples with out-of-reference mutational signatures

Various biological and chemical processes leave characteristic patterns, mutational signatures, in the genome. In our recent work, we documented important limitations of the available tools for signature analysis (Medo et al, 2024). We are now working on algorithms specifically designed to overcome the identified shortcomings, in particular, the activity of out-of-reference (unknown) signatures in the analyzed samples. Even when such signatures contribute a small part of all mutations, they can substantially influence the estimation of signature activities, produce numerous false positives, and thus hamper the biological interpretation of the results.

We will employ a combination of analytical and computational approaches to develop approaches to signature analysis that are robust to out-of-reference signatures. Extensive computer simulations will be used to benchmark the developed algorithms over a wide range of conditions. The final results of this project will be implemented as software packages that will improve and simplify the analysis of mutational signatures for the broad community of researchers and other practitioners. This will help to effectively use costly sequencing data and thus enable discoveries and therapeutic decisions that would otherwise be impossible.

Fitting mutational signatures to tumors with known clonal structure

Phylogenetic analysis of genomic data is often used to study the evolution of cancer by dividing the mutations identified in a sample into separate clonal populations and reconstructing their phylogenetic tree. Signature analysis can then uncover biochemical processes that take place in individual clones, thus providing indispensable information about the tumor's evolution. Signature activities can also be estimated for the corresponding bulk sample, and they should, by definition, match closely the sum of estimated signature activities in individual clones. Our results show that this is not the case for many real samples, where clinically relevant mutational signatures are often found only in bulk or in the clones.

To address this problem, we propose to design an algorithm that will jointly estimate signature activities in the clones and the bulk whilst imposing the constraint that bulk signature activities are the sum of clonal signature activities. In this way, we aim to obtain more precise signature estimates for both the clones and the bulk, thus reducing the level of disagreement between these two representations of the analyzed tumor sample. To achieve this goal, we will combine analytical results, numerical optimization, bootstrap uncertainty estimation, machine learning, and benchmarking on a continuously developed model for synthetic mutational catalogs. After extensive testing on synthetic data, the developed algorithms will be employed on mutational catalogs from real samples, which will help to demonstrate their strengths as well as identify and address possible shortcomings. This project is done in collaboration with the International Cancer Genome Consortium, in particular, The Hospital for Sick Children, Toronto, Ontario.

Characterizing the genomic landscapes of metastatic head and neck squamous cell carcinoma (HNSCC) tumors

Lymph node metastases associate with an aggressive disease progression and decreased survival of patients suffering from various solid malignancies. In the case of HNSCC, lymph nodes (LNs) metastasis is a critical clinical manifestation that predicts patient survival. Accordingly, the survival of patients with LN metastases is about half of that of patients without the tumor being spread to the LNs. Although many efforts have been invested in recent years to profile primary HNSCC tumors to enable interventions through precision oncology-based therapies, the genomics of HNSCC LN metastases remain largely unexplored.

In this broad collaborative study that we have initiated a few years ago, we have been employing whole-exome sequencing to profile the genomic landscapes of matched normal, primary, and metastatic tissues from a cohort of HNSCC patients in terms of mutations and copy number variations. The genomic findings are being correlated with clinical parameters, which are available for all patients. Of particular interest are genomic aberrations that correlate with patient responses to various treatment modalities and molecular signatures that may link to and predict various clinical features. Likewise, efforts are in progress to identify and characterize various molecular subtypes of HNSCC that could be subjected to precision-based interventions and to understand phylogenetic processes underlying the evolution of these tumors.

A MET chimeric antigen receptor (CAR) T-cell immunotherapy combined with radiation therapy (RT) in preclinical models of glioblastoma multiforme (GBM)

Despite major investigational efforts, GBM remains a main clinical challenge due to the limited efficacies of the currently used intervention modalities. The MET receptor tyrosine kinase is ubiquitously and aberrantly expressed in GBM. Along with immune checkpoint inhibitors (ICIs), CAR T-based immunotherapy is establishing itself as an important and highly efficient therapeutic strategy. So far, CAR T interventions have been clinically approved for hematopoietic malignancies, however, major attempts are performed to overcome hurdles associated with efficient tumor eradication of solid cancers. In the case of ICIs, remarkable preclinical, as well as clinical results have been reported when combined with RT due to the so-called abscopal effect. Moreover, the mechanisms underlying immune pre-conditioning by RT, which enable the abscopal effect, have been elucidated.

In our studies, we aim to evaluate an anti-MET CAR T strategy along with RT (employing the irradiation-imaging SMART platform) for GBM. The studies use a panel of both long-term as well as GBM stem cells that are being evaluated both in vitro and within orthotopic animal models. Patients-derived xenograft orthotopic models will be established as well.

It is foreseen that additional CAR T-based approaches using antigens other than MET and in combination with RT for additional tumor types, will be established in the near future.

The effect of small cohort sizes and population heterogeneity on differential expression analysis

Differential expression analysis of RNA sequencing data aims to find differences in gene expression in biological samples between two or more experimental conditions (e.g., treated and untreated cells). Albeit statistical methods developed for this task have various means of assessing the statistical significance of the results, we find that this significance is substantially overestimated when analyzing small patient cohorts. We aim to develop methods to: (1) identify the genes whose differential expression generalizes well to other patient cohorts, (2) identify patient subpopulations with distinct differential expression patterns.