Cancer is a multifaceted and heterogeneous disease influenced by various genetic, epigenetic, and environmental factors. It exhibits dynamic changes and inter- and intra-tumor heterogeneity, posing challenges for understanding its functional implications. In my postdoctoral work at NIH, I employed system-based approaches to tackle various challenges associated with cancer, including tumor heterogeneity, epigenetic drug specificity, resistance to targeted therapy, and identifying novel drug targets. Our network-based approach identified key genes mediating BRCA in a subtype-specific manner and key genes associated with therapy resistance in Triple Negative Breast Cancer (TNBC). We further demonstrated the application of the tool in characterizing novel diagnostic, prognostic, and therapeutic candidates in a subtype-specific manner for HPV-positive and HPV-negative head and neck squamous cell carcinoma (HNSCC). Epigenomic drugs, such as HDAC inhibitor (HDACi), show promise but suffer from unpredictability in genomic specificity. Leveraging pre-treatment transcriptome and epigenomic profiles, we developed a machine learning model to predict locus-specific changes in gene expression following HDACi treatment, achieving high accuracy (up to ROC of 0.89). Furthermore, a model trained on one cell line is applicable to another cell line suggesting generalizability of the model.

In this talk, I will present a central limit theorem describing the fluctuations of the number of zeros (local statistics) of various families of L-functions around their mean. The correlations of these fluctuations coincide with those obtained by Wieand, Diaconis, and Evans for the number of eigenvalues of random matrices. The families of L-functions considered here are defined over function fields associated with hyperelliptic curves of large genus over a fixed finite field.

TBA

Algorithms on strings (or sequences or texts) and related data structures are fundamental to computer science and form the backbone of many data processing tools in Computational Biology. Most problems under exact string matching can be solved optimally using classic data structures like suffix trees/arrays. However, they are notorious for their space inefficiency. For example, the suffix tree of a human genome takes about 30-50 times the size of data, which raises the following question. Can we also index the data in a space close to the information-theoretical lower bound? The first breakthrough in this line of research is from early 2000, known as the compressed suffix arrays and the FM-index; they encapsulate the suffix array's string-matching functionality in near-optimal bits. They rely on a technique called Burrows-Wheeler Transform (BWT) and its associated operation called LF mapping. Since then, this has been an active vein of data structure research and has witnessed the applications of BWT-based techniques on more complex string-matching problems in compressed space. To name a few, we have parameterized matching, order isomorphic matching, and string matching on labeled graphs - a problem with tremendous applications in the trending area of bioinformatics called pan-genomics. In this talk, we will see a brief history of this line of research and the details of some elegant algorithmic techniques. BIO: Sharma Thankachan is an associate professor of Computer Science at the North Carolina State University (NCSU), Raleigh, USA. His main research areas include string algorithms and related data structures for indexing such data in highly compressed space. Much of his work deals with the theoretical aspects of problems stemming from applied fields like computational biology and information retrieval. His research is supported primarily by the National Science Foundation (NSF), and he is a recipient of the prestigious NSF CAREER award.