AI-driven drug discovery startup Atomwise raises $45M series A

Artificial intelligence-driven drug discovery has attracted many investments lately. Atomwise, the first company to commercialize deep neural networks for drug discovery, joined the growing list with a $45 million series A.

The size of the round looks normal for a biotech these days, but to company co-founder and CEO Abraham Heifets, it’s a large one for a software company.

Founded in 2012, Atomwise invented a platform called AtomNet, which uses deep learning for structure-based small-molecular binding affinity prediction.

Compared with older versions of machine learning technique, deep neural networks already deliver at- or even better-than-human performance for image recognition, natural language translation and detection of diabetic retinopathies, Heifets told FierceCRO.

“The thing that makes Atomwise different from others using deep neural networks is that … we are looking not only at molecules, but also at the proteins they are trying to hit, and analyze both sides together,” said Heifets.

RELATED: China AI biotech XtalPi lands $15M from Sequoia, Google, Tencent

Incorporating the structure of the protein in its analysis helps provide better accuracy and refine predictions, according to Heifets.  It can not only identify promising compounds from known libraries, but can also make predictions for virtually synthesized novel molecules that have never physically existed.

Heifets said its capability is applied in projects involving hit discovery, lead optimization, off-target toxicity, selectivity, polypharmacology/bispecificity, and cross-species activity.

The company said its system identifies “active” hit compounds at a rate 10,000 times higher than that of physical screens, or 100 times faster than ultra-high-throughput screening, enabling it to screen more than 10 million compounds each day.

Besides an increase in discovery productivity, Atomwise said its software also provides greater understanding of the toxicity, side effects, mechanism of action and efficacy of a drug, far earlier than typical in a drug pipeline, and can therefore help drug developers save cost and reduce risk.

RELATED: BenevolentAI's Proximagen buy creates first AI firm capable of entire drug R&D

So far, AtomNet has been used in 50 projects and is on track to achieve the 100 milestone in 2018, according to Heifets. Those projects span across a wide range of therapeutic areas, including Alzheimer's disease, bacterial infections and antibiotics, nephrology, ophthalmology, immuno-oncology, as well as metabolic and childhood liver diseases. It has relationship with four of the top 10 pharmas, including Merck, as well as with 40 major research institutes.

The new money will be used to grow the team in order to keep up with existing and future demand, said Heifets. As the company’s platform gets larger and more accurate, it also means increased computational costs.

Monsanto Growth Ventures, DCVC (Data Collective), and B Capital Group led the financing round, and was joined by previous investors Y Combinator, Khosla Ventures, and DFJ, along with new investors Baidu Ventures, Tencent, and Dolby Family Ventures.