Molecular science VR startup Nanome launches AI copilot MARA

San Diego-based Nanome, a startup that allows scientists to understand 3D molecular structures in virtual reality, today announced the launch of a generative AI copilot called MARA.

Available through a ChatGPT-like web interface, MARA serves as an assistant of sorts for chemists working in biopharma research and development by providing insights critical to the research process. It can execute routine cheminformatics tasks and provide informative responses to scientific queries, the company said.

The move comes as many continue to raise questions about the potential of AI in critical areas such as drug discovery – and is expected to culminate into a more comprehensive version of MARA that will serve scientists in virtual environments.

“Whether you belong to a prestigious pharmaceutical research and development facility equipped with their own proprietary algorithms or a burgeoning biotech startup seeking a ready-to-use solution, Nanome’s MARA can be tailored to your lab’s unique needs,” Nanome’s CEO Steve McCloskey said in a statement.

What to expect from MARA?

On the surface, MARA works just like OpenAI’s ChatGPT. Scientists launch a text window where they can input their questions.

MARA uses large language models (LLMs) and access to real-time internal data and molecular simulation systems to provide meaningful answers/suggestions in a conversational tone.

“A chemist…could ask a series of questions or hypotheses in which MARA would retrieve data from various real-time or static sources, perform advanced data analysis or simulations, and provide a clear answer alongside observability into the system,” Keita Funakawa, co-founder of Nanome, told VentureBeat. “Over time, MARA will be integrated into Nanome’s XR (AR/VR Platform) with various modalities such as voice, eye tracking, gestures/hand tracking and text-based input.”

Currently, the capabilities of the AI assistant range from exploring, preparing and rectifying various molecular file formats, including molecular/chemistry Structure-Activity Relationship (SAR) data, to performing complex multi-step processes across diverse tools and databases.

This not only gives chemists the power of expert data scientists but also enables them to explore new theories and ideas with AI, accelerating the research process.

“The MARA Platform enables organizations to wire in custom molecular simulation tools, databases, electronic lab notebooks, and more in an easy and secure way. It is also able to support on-premise with open source LLMs such as Llama v2 as well as integrate with off-the-shelf cloud foundation model providers such as OpenAI’s GPT models,” Funakawa noted. 

The co-founder also said that the tool provides hallucination-free answers by using LLMs strictly as a planning and reasoning function of the total system that relies on defined knowledge and deterministic computation and informatics tools to give scientifically accurate information.

“When it isn’t equipped with the proper tool, instead of hallucinating, it will admit that it is not equipped with a particular tool to answer a query. Clear visibility into the reasoning and tools also instills scientific confidence in not just the outcome of the answers but also how the MARA responds,” he said while noting that there are no other systems like MARA and the closest thing would be a fine-tuned foundation model that may still hallucinate.

‘Overwhelmingly positive response’

While the company did not share how many organizations have started using MARA, it did mention that the new offering is being used by leading pharma companies who have been long-time users of the Nanome XR platform. Currently, over half of the global top 20 pharmaceutical companies use the VR technology from the company to better understand molecular structure designs.

“The reception from chemists and even biologists when initially testing MARA has been overwhelmingly positive. One former director of medicinal chemistry at one of the top 5 pharma companies said this could save chemists weeks of time and open new abilities for SAR analysis,” Funakawa added.

Currently, Nanome remains focused on optimizing the experience of MARA for biopharma R&D. However, in the future, the company plans to bring the assistant to its VR platform — as mentioned above — and expand the data it can work with to provide insights. This will include genetics data, clinical data, material science data and data associated with other scientific disciplines.

“Ensuring that the AI is transparent, accountable, and aligns with the needs of the users will be pivotal in gaining trust and widespread adoption,” Funakawa said.