While many proceed to debate the role of AI in healthcare, startups are fully embracing the technology – with the total support of enterprise capital firms. Today based in San Francisco TriomicsA startup searching for to speed up cancer treatment with generative AI announced it has raised $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 huge language models (LLMs) called OncoLLM that streamline the complex and time-consuming oncology-related workflows that medical center employees must complete correct treatment path for a patient.
The models work with a spread of workflow-specific tools and are proven to finish tasks that will normally take days or perhaps weeks in only minutes.
“We have successfully brought together expertise in two complex functional areas: our AI researchers, who specialise in adapting language models to specific domains, and our clinical staff, who’ve many years of oncology-specific experience.” This allows our software to leverage the strengths of those advanced Complementing models while proactively addressing potential deficiencies – all while considering the intricacies of cancer research and care,” Singh said in an announcement.
What exactly does Triomics want to resolve?
Today thousands and thousands of individuals suffer from cancer. The number of latest cases has increased over time and is estimated to affect 35 million people by 2050 – a 77% increase from the 20 million cases in 2022. In this case, medical facilities and cancer treatment centers are inevitably under pressure. primarily because of the shrinking healthcare workforce.
Currently, most cancer care nurses and staff determine patients' treatment pathway or clinical trial eligibility using manual chart reviews, where they manually review your entire longitudinal record to discover relevant data points. This includes every thing from unstructured free-text notes from doctors to check reports and takes lots of time, resulting in clinical delays comparable to patients missing trials or biomarker-driven treatments.
Triomics addresses this problem by making the oncology-focused OncoLLM available to care centers and allowing them to optimize the model using their very own internal datasets to be used with the corporate's workflow automation offerings.
“OncoLLM is actually a family of models, with each model serving different purposes, including retriever and generator models. Some of them are trained from scratch and others are tuned on SOTA open source models.” Our models undergo extensive fine-tuning using each vendor's proprietary data and reinforcement learning, leveraging human feedback for tailored learning. We employ customized models for every partner institution,” Khan told VentureBeat.
Once aligned with the ability, the models are deployed in Triomics' software offerings, which integrate with health system EHRs to support specific care workflows. The company currently has two products: Harmony and Prism. The former curates the info for registration, reporting or research purposes, while the latter handles patient-study matching by pre-screening oncology patients to search out relevant clinical trials. On a big scale, this reduces the time it takes to review patient records from days or perhaps weeks to simply just a few minutes.
When the model and associated software were tested by the Medical College of Wisconsin Cancer Center, teams found that the offering outperformed larger open source and proprietary LLMs in patient-study matching while still using qualified medical examiners and GPT-4 competed much smaller and 35 times cheaper. Since then, the corporate has also developed one other variant of OncoLLM (70B), which surpasses each GPT-4 and medical examiners when it comes to accuracy.
The goal is to attain scale
With this round of funding, Triomics plans to expand its team across functions and expand the reach of the product.
The company has already signed several contracts and goals to draw over a dozen partner institutions by the top of the yr. It says there is no such thing as a fixed pricing strategy because the OncoLLM-based solution is tailor-made for every customer.
“We are either testing or actively collaborating with half a dozen academic medical centers, and the numbers needs to be within the double digits by the top of the summer. We have also begun expanding our customer base beyond academic centers and getting into agreements with large community oncology practices to enhance the lives of as many patients as possible,” said Khan.
While some solutions help align patient studies, Khan points out that the corporate has developed a specialization in oncology with OncoLLM-based software. Additionally, he says most other solutions on this space usually are not Gen-AI native and depend on leveraging/modifying older technologies without the scaling advantage or incremental ROI that the industry demands.