Automated Life Science A.I. Generation System Unveiled

(Gerd Altmann, Pixabay. https://pixabay.com/illustrations/dna-analysis-research-7237234/)

21 June 2023. A bioengineering lab created a system to make it easier for life scientists to use machine learning for data analysis, without writing their own algorithms. Researchers from the lab of biomedical engineering professor James Collins, at Massachusetts Institute of Technology, also affiliated with the Wyss Institute at Harvard University, describe the system  in today’s issue of the journal Cell Systems.

The research team in data science, engineering, and biology seek to ease the use of artificial intelligence for their biomedical research colleagues. Despite the technical sophistication of life scientists, many researchers have neither the training nor resources to write, train, and validate their own algorithms. For this task, the MIT-Wyss Institute team developed a system called BioAutoMated to help researchers readily identify and apply effective A.I. models to their life science data.

The authors say BioAutoMated aims to build on current automated machine learning or autoML tools. “Our tool is for folks who don’t have the ability to build their own custom ML models,” says MIT engineering graduate student and co-lead author Jacqueline Valeri in a Wyss Institute statement, adding “We wanted to make it easy for biologists and experts in other domains to use the power of ML and autoML to answer fundamental questions and help uncover biology that means something.” Valeri and colleagues make BioAutoMated code available on the GitHub software platform.

The authors note that most autoML tools today are built around one type of algorithm. The team says BioAutoMated models data sets against three types of algorithms for output review and selection by the researcher: neural networks often used for machine learning, swarm intelligence reflecting decentralized and self-organized behavior, and Tree-based Pipeline Optimization Tool or TPOT models often used in genetic programming.

Algorithm in 26 minutes and 10 lines of code

In addition, BioAutoMated is designed to take life science input data directly with little preprocessing needed, an improvement over today’s autoML systems, according to the authors. As a result, says the team, BioAutoMated can directly process data for DNA and RNA sequences of any length, amino acids, and glycan sugar molecules found on cell surfaces. And from those data, BioAutoMated constructs models for predicting biological functions.

The authors tested BioAutoMated on two sets of research tasks. In one case, BioAutoMated evaluated changes in RNA sequences on the efficiency of a ribosome, where cells synthesize proteins, to translate that RNA into an E. coli bacterial protein. The team says BioAutoMated identified a swarm intelligence model that generated an algorithm to accurately predict the outcome, in about 26 minutes and with 10 lines of code.

In another exercise, the team provided BioAutoMated with data for algorithms to answer specific questions about the efficacy of peptides and glycan sugars under various conditions. One algorithm, say the authors, identified amino acids in peptides that accurately predict binding of antibodies to the macular degeneration drug ranibizumab. And in another test, BioAutoMated devised an algorithm to classify the ability of various glycans to produce an immune response, with the results used to refine related work on nanoscale synthetic gene circuits called toehold switches that provide precise control over gene expression, in this case detecting RNA from the Zika virus.

“Machine learning and artificial intelligence tools have been around for a while now, but it’s only with the recent development of user-friendly interfaces that they’ve exploded in popularity, as in the case of ChatGPT,” notes Collins. “We hope that BioAutoMated can enable the next generation of biologists to faster and more easily discover the underpinnings of life.”

More from Science & Enterprise:

• Primate Genome A.I. Model IDs Human Disease Variants
• AI-Designed Cancer Vaccine Shown to Act Against Melanoma
• Rare Disease Search Adds Large Language Model A.I.
• Real World Evidence Shown to Emulate Some Clinical Trials
• $50M Challenge Seeks Quantum Life Science Applications

We designed Science & Enterprise for busy readers including investors, researchers, entrepreneurs, and students. Except for a narrow cookies and privacy strip for first-time visitors, we have no pop-ups blocking the entire page, nor distracting animated GIF graphics. If you want to subscribe for daily email alerts, you can do that here, or find the link in the upper left-hand corner of the desktop page. The site is free, with no paywall. But, of course, donations are gratefully accepted.

*     *     *