Leading research at the intersection of artificial intelligence and biomedical sciences. Specializing in protein design, molecular modeling, and therapeutic discovery using cutting-edge machine learning approaches to solve complex biological challenges.
As a dedicated Machine Learning Researcher, I specialize in advancing drug discovery and protein engineering through innovative computational methodologies that integrate artificial intelligence with biological systems. Leveraging extensive expertise in deep learning, bioinformatics, and computational chemistry, I develop novel algorithms for protein design, molecular optimization, and therapeutic target identification, addressing critical challenges in biomedical research. My work is driven by a commitment to harnessing the transformative potential of AI-biology convergence to accelerate drug discovery and enhance its accessibility, ultimately contributing to global health solutions. Through interdisciplinary collaborations, peer-reviewed publications, and open-source contributions, I aim to advance the scientific community’s understanding of AI applications in modern medicine.
Developed MetaLLM, a transformer-based model for protein-metal binding prediction, achieving 15% improvement over state-of-the-art. Built parameter-efficient fine-tuning frameworks for protein language models using LoRA and adapters. Created interpretable ML models for antibody engineering and personalized therapeutics. Led cross-disciplinary collaboration with researchers on American Heart Foundation (AHA)-funded projects.
Led undergraduate Python programming labs, facilitating hands-on learning and mentoring students in programming concepts and best practices.
Designed privacy-preserving ECG analysis models to monitor cardiac health using a Variational Autoencoder framework, enhancing patient data security and model interpretability.
Led development of production-scale FaceAI engine, implementing anti-spoofing algorithms using ensemble deep learning models (ResNet, EfficientNet). Optimized inference pipeline, reducing latency by 20% through model quantization and TensorRT.
Built computer vision pipeline using YOLOv3 for real-time asset detection with 95% accuracy. Developed time-series forecasting models for business intelligence using LSTM and Transformer architectures. Deployed scalable ML services on AWS (EC2, Lambda, SageMaker) handling 10K+ concurrent users.
Designed data-centric business policies, increasing company online presence by 25%. Conducted market research to position software products, integrating recommendation systems for enhanced brand placement.
Deep expertise in developing and deploying machine learning models for biomedical applications, with focus on transformer architectures and graph neural networks.
Comprehensive knowledge of biological systems and computational approaches to analyze complex biological data and design novel therapeutic solutions.
Specialized in computational approaches to drug design, molecular optimization, and therapeutic target identification using state-of-the-art algorithms.
Strong foundation in statistical analysis, data visualization, and large-scale data processing for biological and clinical datasets.
Developed a comprehensive platform that integrates protein language models with structural biology data to design optimized antibody therapeutics. The system combines ESM-2 and ProtGPT2 models with custom fine-tuning strategies, achieving significant improvements in binding affinity prediction and reducing experimental validation time by 60%.
Created an advanced machine learning pipeline that combines chemical representations (SMILES, molecular fingerprints) with physicochemical descriptors to predict drug-induced toxicity. The system uses ChemBERT for chemical understanding and achieves 92% accuracy in cardiotoxicity prediction, significantly outperforming traditional QSAR approaches.
Developed a sophisticated deep learning system for analyzing 12-lead ECG signals using a novel CNN-VAE architecture. The system addresses demographic and acquisition bias in cardiac risk prediction, achieving 88% sensitivity and 94% specificity across diverse patient populations while providing interpretable clinical insights.
Built a comprehensive framework for molecular property prediction using graph neural networks. The system implements multiple GNN architectures with novel attention mechanisms, establishing state-of-the-art performance on 12 benchmark datasets and providing insights into optimal architectural choices for molecular learning tasks.
Developed an integrated suite of tools for predicting protein function from sequence and structural data. The platform combines multiple machine learning approaches including transformer models, structural analysis, and evolutionary information to provide comprehensive functional annotations with confidence scores and biological interpretations.
Created a comprehensive platform for analyzing large-scale clinical datasets with focus on identifying biomarkers and predicting patient outcomes. The system integrates multiple data modalities including genomics, proteomics, and clinical records, providing automated preprocessing, feature selection, and predictive modeling capabilities.
Interested in collaborative research, consulting opportunities, or discussing innovative projects in AI and computational biology? I'm always open to meaningful partnerships that can drive scientific discovery forward.
shishir@ku.edu
EECS
Research Division
Lawrence, Kansas
Open to Remote Collaboration