US-based biotech company Atomic AI has introduced ATOM-1, a propriety platform component, representing a large language model (LLM) that leverages chemical mapping data to predict the structure and function of ribonucleic acid (RNA).
The goal of the model is to address challenges associated with designing RNA therapeutics by providing a tool that can optimise key characteristics of RNA modalities such as stability, toxicity, and translational efficiency.
Researchers at Atomic AI collected large-scale chemical mapping data, including millions of RNA sequences and more than a billion nucleotide-level measurements, using custom wet-lab assays. ATOM-1 is trained on this dataset.
In the announcement accompanying the launch, Atomic AI founder and CEO Raphael Townshend said: “Through machine learning and generative AI, we now have a unique opportunity with ATOM-1 to predict RNA structure and function with high precision by tuning it with just a small amount of initial data points.”
The announcement comes as there are questions surrounding the data accuracy of generative AI platforms. A recent paper published in JAMA Ophthalmology found that GPT-4, the latest edition of OpenAI’s LLM, is capable of generating false datasets. The AI can use a set of parameters and produce semi-random datasets when prompted to find data that supports a particular conclusion.
Artificial intelligence (AI) is a strong fixture in the healthcare industry. According to a report on GlobalData’s Medical Intelligence Center, the market for AI platforms for the entire healthcare industry will reach $4.3bn by 2024. GlobalData forecasts that the entire global AI market will be worth $383.3bn in 2030.
In April 2023, Google announced the limited access of its LLM, Med-PaLM 2, to be used by the healthcare sector to answer medical questions accurately and safely. The technology was evaluated against clinician-backed parameters, including medical reasoning, scientific consensus, bias, and likelihood of possible harm.
Google reported that the platform performed firmly on medical exam-style questions, but development is ongoing.
