Rare diseases are difficult to diagnose due to limitations in traditional genomic sequencing. Wolfgang Pernice, assistant professor at Columbia University, is using AI-powered cellular profiling to bridge these gaps and advance personalized medicine.
At NVIDIA GTC 2024, Pernice shared insights from his lab’s work with diseases like Charcot-Marie-Tooth (CMT) and mitochondrial disorders. His team developed CellNet, an AI-driven system that uses high-resolution images of cells from patients to identify subtle disease-related patterns, for accurate diagnoses and new treatment strategies.
AI at the intersection of genomics and medicine
Genomic medicine has long aimed to translate diagnoses into personalized treatments or cures. While there has been progress, many patients with one of 7,000 known rare genetic disorders still need effective therapies. Pernice’s lab is working toward a faster and more scalable approach to genomic medicine, using computer vision and deep learning to address key barriers in rare disease diagnostics and treatment.
Traditional genetic diagnostics struggle with rare diseases due to the complexity and variability of genetic mutations. Pernice’s lab trains AI models on cellular images using NVIDIA GPUs, including the NVIDIA H100 Tensor Core GPU, with NVIDIA CUDA and cuDNN to identify previously undetectable morphological patterns in patient cells.
Key to their approach is the interventional style transfer (IST) framework, a system that addresses batch effects—variations caused by experimental setups—by generating synthetic datasets that improve the generalizability of AI models. This ensures accurate predictions even with out-of-distribution data, a key step in creating scalable diagnostic tools.
What you’ll learn
This session offers key insights into the transformative potential of AI in rare disease genomic medicine, including:
- Addressing diagnostic gaps in rare diseases: How AI-powered cellular profiling closes diagnostic gaps for conditions like CMT and mitochondrial disorders.
- AI-driven phenotypic discovery: How high-resolution imaging and machine learning uncover disease-associated patterns directly from patient cells.
- Mitigating batch effects with IST: An introduction to the interventional style transfer (IST) framework, which improves model reliability by addressing batch effects and improving generalization across different experimental setups.
- Accelerating drug discovery: How AI accelerates identifying and validating potential drug candidates through precise cellular profiling.
- AI and genomics integration: A look at advanced AI tools, powered by NVIDIA GPUs, changing genomic medicine and driving breakthroughs in personalized healthcare.
Watch the session Computer Vision for Rare Disease Genomic Medicine on NVIDIA On-Demand. Explore more expert-led videos and gain valuable skills by joining the NVIDIA Developer Program.
This content was partially crafted with the assistance of generative AI and LLMs. It underwent careful review and was edited by the NVIDIA Technical Blog team to ensure precision, accuracy, and quality.