Jai Choraria

Bachelor

Staff Scientist

advanced data visualization advanced statistical analysis using R and Python Agile Project Management AI Algorithm Development Algorithms Allele-Specific Expression Analysis Analytic Software Development Analytical thinking API integration Artificial Intelligence in Medicine Artificial Neural Networks AutoML AWS Azure Bash bash scripting Bayesian Analysis Bayesian Modeling Bayesian optimization Bayesian Statistics BERT Bioinformatics Biological Data Analysis Biological Interpretation Biomarker Identification Biomechanics Biomedical Imaging Biomedical informatics and health care Biostatistical Programming Biostatistics BLAST Cancer Bioinformatics Cancer genomics Causal Inference CI/CD CI/CD (Continuous Integration/Deployment) Cloud computing CNN Comparative Genomics Computational Biology Computational Genomics computer vision Conda Cross-functional collaboration Cross-validation Data Curation Data Integration Data Mining Data Pipeline data science Data Visualization Database Management Deconvolution Deep Learning DESeq2 Differential Expression Analysis Dimensionality reduction Disease Modeling DNA Sequencing Docker EEG signal processing ElasticNet Electrophysiology EMG analysis Epidemiology Epigenetics Exome Sequencing Experimental Design FAISS FastAPI Feature engineering Full-Stack Development Functional Genomics gatk GCP GDPR compliance Gene Expression Profiling Generative AI Genome Annotation ggplot2 GitHub GitHub Actions Google Cloud (GCP) GPT Gradient Boosting Hyperparameter tuning ISO 13485 JSON Jupyter Notebooks K-Nearest Neighbors Keras LangChain LLM fine-tuning Machine Learning Machine Learning (ML) Machine Learning models MATLAB Matplotlib Medical device software Multiple sequence alignment Nextflow NumPy Ocular activity detection OpenCV Pandas PCA Phylogenetic Analysis PostgreSQL Python Python (Flask Pytorch R Programming RAG (Retrieval-Augmented Generation) Random Forest REST APIs SAMtools Sarcopenia detection scanpy scikit-image Scikit-learn SciPy scRNA-seq analysis Seaborn Signal Processing Single-Cell RNA-Seq snakemake STAR aligner Support Vector Machines (SVM) t-SNE Tensorflow Transformer models UMAP Variant Calling Wearable sensor data processing Whole-Genome Sequencing XGBoost

Conference Poster(s)

Conducted a comprehensive literature-driven investigation into chimeric antigen receptor (CAR) design strategies for macrophage-based cancer immunotherapy. Systematically evaluated and compared a wide range of candidate CAR constructs to identify the optimal dual-targeting configuration, ultimately selecting a combination of anti-CD47 and anti-HER2 single-chain variable fragment (scFv) binding domains. The rationale centred on simultaneously blocking the "don't eat me" CD47-SIRPα inhibitory checkpoint to restore phagocytic activity, while directing macrophage cytotoxicity toward HER2-overexpressing tumor cells. Researched lipid nanoparticle (LNP) delivery systems modelled on clinical-grade formulations to enable in vivo mRNA transfection of tumor-associated macrophages without ex vivo manipulation. Synthesized findings into a theoretical framework that was presented at the MURC 2025 conference and published as an abstract in the Canadian Undergraduate Research Journal, demonstrating both scientific rigour and effective scientific communication.

Contributed to a machine learning research project focused on endometriosis detection and classification using clinical and imaging data. Designed and implemented a complete end-to-end ML pipeline, beginning with exploratory data analysis (EDA) to identify class imbalances, missing values, and feature distributions. Applied data preprocessing strategies including normalization, one-hot encoding, and SMOTE-based oversampling to address class imbalance inherent to medical datasets. Established and evaluated a suite of baseline models encompassing logistic regression, decision trees, k-nearest neighbours (KNN), and support vector machines (SVM) to benchmark predictive performance across metrics including accuracy, precision, recall, F1-score, and AUC-ROC. Performed systematic hyperparameter tuning using grid search with stratified k-fold cross-validation to optimize model generalization and prevent overfitting on limited clinical data. Investigated ensemble methods including Random Forest and Gradient Boosting (XGBoost, LightGBM) to improve classification robustness. Conducted feature importance analysis using SHAP (SHapley Additive exPlanations) values to identify clinically interpretable predictors, contributing to translational insight into endometriosis biomarkers. Evaluated model calibration and decision threshold optimization to maximize clinical utility, balancing sensitivity and specificity for a high-stakes diagnostic context. Documented all experimental results in structured notebooks using Jupyter, with reproducible pipelines and version-controlled code.

Led the development of a deep learning pipeline for sarcopenia detection and muscle mass quantification from medical imaging data (CT/MRI). Personally contributed to the creation of a custom annotated dataset by manually segmenting and labeling muscle masks across a large cohort of medical images using clinical annotation tools, ensuring high inter-rater reliability through iterative quality checks. Designed and trained multiple convolutional neural network (CNN) architectures — including U-Net and modified ResNet-based segmentation models — on these self-annotated masks to automate skeletal muscle cross-sectional area (CSA) measurement, a key biomarker for sarcopenia diagnosis. Conducted rigorous model benchmarking across architectures by evaluating Dice coefficient, Intersection over Union (IoU), sensitivity, and specificity on held-out validation sets. Performed multi-round hyperparameter optimization using techniques including learning rate scheduling, dropout regularization, batch normalization tuning, and data augmentation strategies (random flipping, rotation, elastic deformation) to reduce overfitting and improve generalization across patient demographics. Achieved a dramatic reduction in radiologist processing time — from approximately 30 minutes per scan down to 0.1 seconds — by replacing manual segmentation workflows with the trained automated model, representing a clinically significant efficiency gain. Integrated the inference pipeline into a deployable format to support radiologist workflow integration, enabling scalable sarcopenia screening in clinical environments. Contributed to the technical documentation and validation of model performance for potential translational and regulatory consideration.