NEW YORK – Bioinformatics firm InSilico Medicine is broadening its reach with new products aimed at making the company's multiomic analysis technology accessible to more potential users.
The Hong Kong-based company recently launched an open-source large language model (LLM) called Precious3GPT aimed at both drug discovery and deriving molecular aging clocks, as well as PandaOmics Box –– a more secure version of its PandaOmics multiomic data analysis software meant for hospitals and other users who need to store sensitive personal data.
Precious3GPT is classified as a transformer, meaning that it is a type of neural network that learns context and thus meaning by tracking relationships in sequential data, such as the words in a sentence.
The model incorporates longitudinal methylation, transcriptomic, and proteomic data from multiple tissue types and cell lines across four species: mouse, rat, monkey, and human. The model analyzes this data to identify drugs that might affect the aging process, predict drug responses, and derive molecular aging clocks for individuals.
"Now we can create life models that understand how different life forms react to different chemicals," said CEO and Cofounder Alex Zhavoronkov, adding that this enables InSilico to go "way beyond" drug discovery.
Zhavoronkov said that Precious3GPT features a more efficient training process that requires fewer steps to achieve performance that is on par with larger models. A study published on BioRxiv showed that Precious3GPT outperformed the much larger BioGPT-Large and OpenBioLLM-8B models in tests involving biological function and cell localization gene annotations.
Precious3GPT's efficiency and multimodality are what sets it apart from other tools, Zhavoronkov said.
Although not widely commercialized, the idea of deriving and using molecular clocks to better understand how diseases arise and evolve, to identify therapeutic targets, and to design therapeutic molecules is an area of active research.
Last year, for example, researchers from Stanford University and Aarhus University published a proteomics-based molecular clock approach to tracking aging, eye diseases, and conditions such as Parkinson's disease from eye fluid samples.
Another group from St. Jude Children's Research Hospital identified faster epigenetic aging patterns among pediatric cancer survivors, which correlated with higher risks of developing obesity.
"Precious3GPT … is unique compared to most aging clocks that work with a single data modality, such as DNA methylation," said Jesse Poganik, an instructor in medicine at Harvard University, whose research focuses on the biology of aging.
"We know that different clocks capture different signals," Poganik said, "but we are working to understand and define the use cases for different clocks. This is because different clocks may be most ideally suited to different uses, [such as] evaluation of biological age, prediction of disease, evaluation of response to interventions, etc. It will be interesting to see how Precious3GPT and other advanced models will help us to refine our understanding of aging biomarkers."
Poganik also serves on the executive committee of the Biomarkers of Aging Consortium and said that the consortium "will likely" launch a study to compare the performance of Precious3GPT against existing biomarkers of aging in the near future.
Zhavoronkov said that InSilico focuses on aging research because of the impact it can have in drug development and because this kind of research helps train artificial intelligence (AI) models to better understand how biology acts with time.
"Think of it as training AI not on pictures of life but on video from birth to death," he said. "If your AI does not understand the process of change in biology during aging it will not do well in new target discovery."
InSilico has made Precious3GPT open source, and Zhavoronkov said that the company intends to keep it that way. "The business model [is to] put it in the wild for other people to use, to see where it can be used, and also to validate it massively," Zhavoronkov said. "It's just such a wonderful tool; it would be a crime for me to not open source it."
PandaOmics Box
InSilico is also rolling out PandaOmics Box, a self-contained device that enables use of the company's PandaOmics target discovery platform without an internet connection, thereby keeping sensitive patient information safe from data breaches. This enables InSilico to market PandaOmics to hospitals and other organizations with stringent personal data security needs.
"The Box doesn't even have Wi-Fi, so it provides you with ultimate protection because it's 100 percent disconnected from the internet," Zhavoronkov said. "Once you give it to the hospital, it's theirs, and all the computing is done on their premises."
Zhavoronkov said that regular updates are provided to users, who install it themselves at their respective facilities.
PandaOmics Box features an open API that allows users to customize its workflows for their individual needs and conduct in-house therapeutic target discovery.
InSilico has so far deployed PandaOmics Boxes to two "major hospitals" as part of a pilot program, but Zhavoronkov said that he couldn't yet say which ones. The company hopes to learn from this program how best to position the new platform for wider commercialization planned for later this year or the beginning of next year.
"We're currently designing the commercial strategy for how to [sell] it properly in different geographies," Zhavoronkov said.
Zhavoronkov said that requests from hospitals to use PandaOmics drove the development of PandaOmics Box.
"We had to turn their requests down because our policy [is to] not work with nonpublic patient data," Zhavoronkov said. "We decided to instead to provide them with hardware that does not access the internet and has all software locally."
Zhavoronkov said that the total market size for PandaOmics Box is difficult to know but estimates that there may be "a couple hundred" hospitals within the US, "a hundred or more" in the Middle East and North Africa, and potentially thousands across China and India that are potential customers.
"We think that since many of the doctors want to publish their research and hospitals would want to discover new targets, the market may be quite substantial," he said.
While InSilico Medicine works toward full commercialization of its new products and continues to classify itself as a drug discovery company, it has also been applying its technology to areas beyond biomedicine, such as agriculture and environmental sustainability.
While perhaps not directly related to the company's core business of aging research and drug discovery, Zhavoronkov said that each branch of the company's business contributes data that helps the company build and improve its biological modeling tools. "In agriculture, we get a lot of data on toxicity and environmental impact on health," Zhavoronkov noted.
InSilico Medicine currently has facilities in the US, China, Taiwan, and the United Arab Emirates and aims to have a global reach. "The more groups use our software," Zhavoronkov said, "the better it becomes."