Gautham Krishna

– Built an end-to-end transformer (ESM and fine-tuned ESM) pipeline for KRAS variant effects, deriving structure-aware embeddings and features, using manifold learning and supervised models with uncertainty-weighted loss to predict ddPCA energy phenotypes and surface allosteric/druggable pockets

Developed novel cross-attention-based ML methods to improve Protein-Protein Interaction prediction using Protein Language Models(ESM2, Prot-T5, Prot-Bert) with Layerwise Relevance Propagation (explainable AI) strategies to capture interpretability in attention networks.

– Integrated GPT-4 into ZoneQuest AI, enabling RAG-based zoning analysis, structured data visualization, and document summarization across 10 plus U.S. cities, enhancing accessibility for 100 plus users. – Developed Python and MySQL workflows for structured data conversion of zoning data, enhancing ML integration, AI-driven summarization, and knowledge retrieval, improving efficiency by 20% for urban planning insights.

Dr. Amirali Aghazadeh Mohandesi's Lab, School of Electrical and Computer Science Project Title: ProtiGeno: a Prokaryotic Short Gene Finder using Protein Language Models • Contributed to the development of a Deep Learning Framework: ProtiGeno specifically targeting short prokaryotic genes using a protein language model (ESM-1b)and training a Neural Network Classifier to annotate these genes. • Collected and preprocessed short coding and non-coding data used for training ProtiGeno and performed 10-fold cross-validation to test its performance. • Utilized GeneMarkS, Prodigal, Balrog, and other ML models to benchmark our performance concerning the short gene annotation. • Conducted structural analysis utilizing AlphaFold2 and ESMFold to investigate the False positives and gain insights into the predictive features.

• Developed Tool Augmented Language Models (TALM) as a chemistry toolkit, integrating OpenAI APIs and HuggingFace to streamline tasks across organic synthesis, drug discovery, and materials design. • Leveraged advanced language models, including GPT-4, Mistral, Gemma, Falcon , and Llama2, by utilizing LangChain’s implementation of custom agents. • Benchmarked performance for unprompted models, where Mistral and Gemma's model demonstrated the best compromise between success rate (0.6) and accuracy (0.76).

Dr. Juan Marcos Castillo's Lab , School of Computer Science -Project Title: IoT-Cybersecurity using Machine Learning for Distributed Denial of Service(DDoS) Attacks • Implemented Classification Algorithms like Decision Trees, Linear Regression, Random Forest to evaluate the IOT CICDDoS2019 dataset on DDOS attacks • Determined that the Linear Regression model outperformed the other models with an accuracy of 99.8 with 2 weighted features restricted as labels.

Dr. Ragothaman Yennamalli Lab, School of Chemical and Biotechnology -Project title : 1. Coexpression And Meta-Analysis of Genes Involved in Lignocellulosic Deconstruction 2. Long time-scaled structural dynamics-based mutation analysis of GNE Myopathy variants in the Indian Subcontinent