Ramana Davuluri PhD

Ramana V. Davuluri's research focuses on the development of machine-learning algorithms and informatics solutions for problems in isoform-level gene regulation and precision medicine. Using state-of-art machine learning and statistical methods, the Davuluri group has been developing predictive algorithms for TF binding sites, Pol-II promoters, and transcriptional modules from ChIP-seq/chip data, and Isoform level gene expression estimation and exon-level differential expression analysis from exon-array and NGS data.

His group has developed a PIGExClass (platform-independent isoform-level gene-expression-based classification-system) data-mining framework, a robust computational approach to derive and then transfer gene signatures from one analytical platform to another for designing clinically adaptable molecular subtyping assays. PIGExClass was successfully applied to glioblastoma and high-grade ovarian cancer for developing robust platform-independent molecular profiling assays for cancer patient stratification. The classifiers derived by PIGExClass have the potential to develop into prognostic biomarkers for the stratification of cancer patients.

Davuluri's group has developed an informatics pipeline to evaluate whether some protein isoforms (resulting from alternative splicing) have the potential to serve as therapeutic targets in cancer, which is a missing piece in the majority of drug discovery processes. By integrating information from publicly available databases, his group has curated FDA-approved or investigational-stage small-molecule cancer drugs that target different genes. By analyzing the interactions with binding pocket information, his group has found that 76% of drugs either miss a potential target isoform or target other isoforms with varied expression in multiple normal tissues. His group is investigating isoform-level drug-target interactions that could play an important role in on- and off-target effects at the splice-variant level to enhance the productivity of drug-discovery research.

Davuluri's group has recently developed a novel pre-trained bidirectional encoder representation, named DNABERT, to capture a global and transferrable understanding of genomic DNA sequences based on up and downstream nucleotide contexts. The significance of his work is the creation of a pre-trained DNABERT model and fine-tuning modules for specific sequence prediction tasks, based on a new breed of deep-learning algorithms. These algorithms optimize the specific sequence prediction tasks and interpretation of germline and somatic sequence variants in non-coding genomic regions that play an important role in cancer initiation and progression. His ongoing research is focused on developing novel computational tools for predicting polysemous cis-regulatory elements and splice sites that are disrupted in cancer genomes, which will likely lead to the identification of non-coding genetic variants that offer crucial applications in cancer biomarker discovery.