Triomics taps LLMs to accelerate cancer care, raises $15M

Join us in returning to NYC on June 5th to collaborate with executive leaders in exploring comprehensive methods for auditing AI models regarding bias, performance, and ethical compliance across diverse organizations. Find out how you can attend here.

As many continue to debate the role of AI in healthcare, startups are going all in on the technology – with the full support of venture capital firms. Today, San Francisco-based Triomics, a startup looking to accelerate cancer care with generative AI, announced its $15 million from Lightspeed, Nexus Venture Partners, General Catalyst and Y Combinator. 

Founded by former MIT and Adobe researchers Sarim Khan and Hrituraj Singh, Triomics has developed a family of large language models (LLMs), dubbed OncoLLM, that streamlines the complex and time-consuming oncology-related workflows staffers at medical centers have to go through to determine the right treatment path for a patient. 

The models work with a set of workflow-specific tools and have been proven to do tasks that would usually take days or weeks in just minutes.

“We have successfully merged expertise in two complex functional areas: our AI researchers, who are specialized in customizing language models to specific domains, and our clinical staff, who have decades of oncology-specific experience. As a result, our software can complement the strengths of these advanced models while also proactively addressing potential flaws, all with the intricacies of cancer research and care in mind,” Singh said in a statement.

VB Event

The AI Impact Tour: The AI Audit

Join us as we return to NYC on June 5th to engage with top executive leaders, delving into strategies for auditing AI models to ensure fairness, optimal performance, and ethical compliance across diverse organizations. Secure your attendance for this exclusive invite-only event.

Request an invite

Triomics software

What exactly Triomics aims to solve?

Today, millions of people suffer from cancer. The number of new cases has been increasing over the years and is estimated to touch 35 million people by 2050 – a 77% increase from the 20 million cases in 2022. In this situation, medical and cancer care centers are bound to be under pressure, especially due to the dwindling healthcare workforce.

Currently, most nurses and cancer care staffers determine patients’ care pathway or clinical trial eligibility with manual chart reviews, where they sift through the entire longitudinal record manually to identify relevant data points. This covers everything, from doctors’ unstructured free-text notes to test reports, and takes a lot of time, leading to clinical delays such as patients missing out on trials or biomarker-driven treatments. 

Triomics tackles this problem by providing care centers with Oncology-focused OncoLLM and allowing them to fine-tune the model – using their own internal datasets – for deployment with the company’s workflow automation offerings. 

“OncoLLM is essentially a family of models, each model serving different purposes including retriever and generator models, some of them are trained from scratch and some are fine-tuned on SOTA open-source models. Our models undergo extensive fine-tuning on each provider’s proprietary data and reinforcement learning, leveraging human feedback for tailored learning. We deploy customized models for each partner institution,” Khan told VentureBeat.

When the models are institution-tuned, they are deployed across Triomics’ software offerings that integrate with health system EHRs to help with specific care workflows. Currently, the company has two products in place: Harmony and Prism. The former curates the data for registry, reporting or research needs, while the latter handles patient-trial matching by prescreening oncology patients to find relevant clinical trials. At scale, this cuts the time to review patient charts from days or weeks to mere minutes.

When the model, and the associated software, were tested by the Medical College of Wisconsin Cancer Center, the teams found that the offering outperformed larger open-source and proprietary LLMs at patient-trial matching and rivaled qualified medical experts and GPT-4, despite being much smaller and 35 times less expensive. Since then, the company has also developed another variant of OncoLLM (70B) that surpasses both GPT-4 and medical experts in terms of accuracy.

Performance of OncoLLM
Performance of OncoLLM

Goal to hit scale

With this round of funding, Triomics plans to increase its team across domains and scale up the reach of the product. 

The company has already signed a few deals and is targeting to rope in over a dozen partner institutions by the end of the year. It says there’s no fixed pricing strategy as the OncoLLM-based solution is tailored for each customer.

“We are either piloting or actively working with about half a dozen academic medical centers, which should be double digits by summer’s end. We also have started to expand our customer base beyond the academic centers entering into agreements with large community oncology practices, to make an impact on as many patient lives as we possibly can,” Khan said.

While some solutions help with patient-trial matching, Khan notes that the company has developed a specialization around oncology with OncoLLM-powered software. Additionally, he says most other solutions in this space are not gen AI native and are reliant upon utilizing / modifying legacy technologies without the scale benefit or step-function ROI the industry is asking for.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top