How Llama helps Biofy Technologies in the fight against antibiotic resistance

Antibiotic resistance poses a significant threat to global health and is a common cause of death in hospitals. To address this challenge in their home country of Brazil, the biotech company Biofy Technologies has developed a groundbreaking platform using Llama that reduces diagnostic time for antibiotic resistance from five days to less than four hours.

Biofy developed a system to identify bacteria and their antibiotic resistance using a vector database, where each piece of bacterial DNA is linked to an entry in the database. The original system was built using the NCBI genome database, which has around 720,000 samples of bacterial DNA. Biofy needed to add more DNA variations because of how frequently bacterial DNA can change.

To do that, the Biofy team customized the Llama 3.2 90B model to work with the original genome database and generate new synthetic DNA samples. The result was Biofy’s Abby Recommender. The solution runs on Oracle Cloud Infrastructure (OCI) and dedicated GPUs. Abby Recommender can analyze DNA, identify bacteria, and recommend the correct antibiotic to use based on bacteria resistance.

“A platform that can rapidly identify antibiotic-resistant bacteria and propose effective treatment can save lives,” says Biofy CEO Paulo Perez. “Llama is a fundamental component used to create the genome vector database for our solution. We chose it because it is open source, fully customizable, and suited to the problem we needed to solve.”

Biofy uses Llama with OCI Generative AI and Oracle AI Vector Search. The team developed a proprietary model to convert DNA using Oracle Autonomous 23ai Database with Vector DB.

An open source approach was important to the Biofy team because they needed to freely adapt the model to their requirements for creating DNA synthetic data.

“It would have taken years of investment in R&D to achieve a similar solution,” Perez says.

Biofy’s algorithms check whether a genome is valid and stable for bacterial DNA. The team then tested the synthetic data created by the Llama 3.2 90B model.

“It’s clear that we can use AI beyond traditional applications like chatbots and for sales solutions,” says Perez. “We can use it to help save lives and have a better quality of life.”