Biofy Accelerates Lifesaving Disease Diagnosis, Powered by NVIDIA

To decrease the testing-to-treatment timeline, Biofy implemented a cloud-native pipeline on Oracle Cloud Infrastructure (OCI) that runs NVIDIA-accelerated algorithms to assemble full bacterial genomes, embed them as vectors, and match these vectors against a larger database of microbial genomes and resistance markers.

This process is followed up by Biofy's recommendation engine, named Abby, that is able to analyze identified resistance genes and suggest the most effective antibiotics or treatment methods for specific infections. Abby's results are seamlessly fed into the clinical workflow to ensure specified plans are quickly put into action—saving time and lives.

During the development process, Biofy experienced a key bottleneck: high-throughput genome assembly, vector search across hundreds of thousands of genomes, and large-model training all require massive parallel compute that CPUs simply cannot provide within ideal clinical time windows.

Biofy utilized NVIDIA GPUs, combined with OCI AI infrastructure, to offer high-performance AI training and inference—50% better at up to half the cost—with tight, streamlined integration using Oracle's data and vector platforms.

The end-to-end diagnostic process is now done in about four hours, versus five days, for standard lab results—a 30x speedup. Additionally, instead of focusing on a fixed panel of genes, Biofy's solution considers complete genomes and resistance patterns, increasing diagnostic accuracy to around 95% in comparison to traditional methods.

Lab scientists and infectious disease specialists can now gain rapid, high-confidence results that fit their existing workflows, reducing diagnostic ambiguity and burnout.

Biofy's leading biology model accounts for ongoing variability in infectious disease detection and treatment. Based on the Llama 3 architecture, the model continuously generates 100,000+ synthetic DNA sequences to anticipate new resistance variants and keep accuracy high as bacteria mutates.

NVIDIA's full-stack platform was utilized to create Biofy's system, including NVIDIA RAPIDS™ to accelerate large-scale DNA matrix operations and embeddings on GPUs, NVIDIA® CUDA®-accelerated libraries via OCI, and 32x H100 Tensor Core GPUs for training and serving LLM-based and proprietary genomic models. RAPIDS includes NVIDIA cuDF for processing large volumes of DNA data and NVIDIA cuML for correcting errors present in DNA.

The storing of genomic vectors was done in Oracle Autonomous AI Database with Oracle AI Vector Search to keep model inference fast and cost efficient. Oracle is evaluating NVIDIA cuVS to accelerate vector search and index builds on Oracle Database 26ai.

This automated, NVIDIA-accelerated AI pipeline can transform the current system to be a streamlined process that runs hundreds of thousands of bacterial genomes and synthetic DNA variations in minutes, allowing lab teams and clinicians to focus on patient care and treatment plans instead of manual analysis.

Since the commercial launch of Biofy's solution, several thousand tests have been performed, and more than a hundred lives have been saved. New cases emerge daily, especially cases where labs are unable to identify bacteria, and Biofy provides an accurate result in a few hours.

A recent case reported a patient with a serious leg infection, who was at risk of amputation due to the laboratory having a difficult time identifying the bacteria fast enough to implement a care plan within the viable treatment window. Using Biofy's Abby, it was possible to diagnose and start the best-fit medical plan in the span of a few hours, saving this patient's leg and life.

Currently, Biofy is working to expand its genomic platform to new hospitals in Brazil and to establish a new entity in the U.S. to serve the American and global markets. Partnerships are underway, such as with Oracle's healthcare division and several American educational and research institutions.

The team is building a large peptide foundation model to design novel antibacterial peptides for pathogens that are resistant to current treatments—aiming to deliver first-class, AI-generated therapeutics over the next two years.

As part of its technology evolution strategy, Biofy plans to expand its current platform to support the identification and analysis of fungi and viruses, broadening its scope in infectious diseases and increasing diagnostic accuracy and clinical impact through advanced AI, with the ongoing goal of saving lives globally.