Easy Cash4ads»
Last week, a group of Amazon scientists and engineers gathered to dream big. The event was all about machine learning, a powerful type of artificial intelligence that has already transformed Amazon’s business and those of other tech giants.
What was different about this AI conclave was its focus on audacious possibilities in the medical field, such as using ML to revolutionize drug discovery, clinical trials, genomics and related areas.
Insider obtained documents that reveal the topics, goals and challenges discussed. Together, they show Amazon’s ambition to take on Google’s DeepMind, a pioneer in AI-powered scientific discovery. This could take Amazon from dabbling in healthcare services, and turn it into a potentially serious player in the future of medicine.
“The demarcation line between core Amazon/AWS business and life science and healthcare is shifting,” said Amazon scientist and senior solutions architect Sergey Menis, according to a transcript of his comments seen by Insider. “We are increasingly more specialized in healthcare and life sciences.” An Amazon spokesperson declined to comment.
Menis developed a nanoparticle that underpins a promising HIV vaccine candidate. He was joined at last week’s Amazon Machine Learning Conference by Amazon’s chief medical officer Taha Kass-Hout.
One of the workshops was about machine learning for “human health.” A related discussion panel focused on “ML for Biomolecules & Biosystems in the real world.”
“In this workshop, we focus on three strategic topics where ML has the potential for continued specific impact: multi-omics, drug discovery, and medical diagnostics,” one of the documents said. Multi-omics pulls data from genomics and similar fields including the study of the human microbiome.
Kass-Hout expressed interest in areas such as next-generation data sequencing for genomics, proteins, and small molecules for drug discovery. In an internal post shared prior to the event, the executive called on Amazon employees to submit research papers up to 6-pages long that could be shared and discussed during the workshop covering the following topics:
During a separate discussion panel, Menis delved deeper into machine learning and healthcare. When asked to share the “most exciting thing” at Amazon, Menis mentioned the blurring lines between the company’s healthcare business and its research efforts, pointing to Amazon’s Diagnostics and COVID testing initiatives as well as the Amazon Genomics CLI tool.
Immunogenicity, or a substance’s ability to produce an immune response, is a problem Menis advised others to focus on, as it’s important for vaccines and protein therapeutics.
Therapeutic development in the “digital world” is an area that remains relatively unexplored, he added. The biggest advancement in machine learning and therapeutics would occur when “an accurate digital twin of a human and their environment” is created, he said.
“Think of running a large compute job on AWS instead of running a large clinical trial across the world. That would be truly revolutionary!” Menis said.
Menis mentioned DeepMind’s AlphaFold several times, and urged his colleagues to go beyond its capabilities.
AlphaFold is an AI program developed by DeepMind, a research lab owned by Google parent Alphabet. It can accurately predict and map out protein structures, an important step in developing new medicines. Earlier this year, AlphaFold published the predicted structure of over 200 million proteins, which DeepMind said essentially covers the “entire protein universe” and called the start of “a new era of digital biology.”
Menis said the next trend is “protein hallucination,” where AI models “fill in the blanks and develop a solution.”
AlphaFold essentially uses ML to produce a digital representation of millions of proteins. Menis said that approach can be taken in other parts of the drug development and testing process.
“Pick a technique in the therapeutic development pipeline and pose the question – “How can this be represented in the digital world?” he said. “Replacing the expensive equipment in the lab with a computational algorithm democratizes access for researchers worldwide.”
The biggest challenge with all this may be Amazon’s stomach for long-term expensive projects. When asked to share the “biggest blocker” for the success of machine learning in medicine and biotech, Menis picked “patience and performance.” The development process could drag on and the software quality bar is higher than other applications, which requires a longer timeline, he said.
“Commitment from leadership to a long-term strategy. Computational work is fast. Biology and life sciences in general takes time,” Menis said. “Validating solutions takes longer than what Amazon/AWS are used to.”
Still, Amazon CEO Andy Jassy has identified healthcare as a major new priority, so Menis and his colleagues may have support from the top.
Do you work at Amazon? Got a tip? Contact reporter Eugene Kim via the encrypted messaging apps Signal or Telegram (+1-650-942-3061) or email (ekim@insider.com).
Read next
Listen to The Refresh, Insider’s real-time news show
Listen to The Refresh, Insider’s real-time news show
Read next