黑料吃瓜资源

The human genome is three billion letters of code, and each person has millions of variations. While no human can realistically sift through all that code, computers can. Artificial intelligence (AI) programs can find patterns in the genome related to disease much faster than humans can. They also spot things that humans miss. Someday, AI-powered genome readers may even be able to predict the incidence of diseases from cancer to the common cold. Unfortunately, AI鈥檚 recent popularity surge has led to a bottleneck in innovation.

鈥淚t鈥檚 like the Wild West right now. Everyone鈥檚 just doing whatever the hell they want,鈥� says 黑料吃瓜资源 (CSHL) Assistant Professor Peter Koo. Just like Frankenstein鈥檚 monster was a mix of different parts, AI researchers are constantly building new algorithms from various sources. And it鈥檚 difficult to judge whether their creations will be good or bad. After all, how can scientists judge 鈥済ood鈥� and 鈥渂ad鈥� when dealing with computations that are beyond human capabilities?

That鈥檚 where GOPHER, the 鈥檚 newest invention, comes in. GOPHER (short for GenOmic Profile-model compreHensive EvaluatoR) is a new method that helps researchers identify the most efficient AI programs to analyze the genome. 鈥淲e created a framework where you can compare the algorithms more systematically,鈥� explains Ziqi Tang, a graduate student in Koo鈥檚 laboratory.

GOPHER judges AI programs on several criteria: how well they learn the biology of our genome, how accurately they predict important patterns and features, their ability to handle background noise, and how interpretable their decisions are. 鈥淎I are these powerful algorithms that are solving questions for us,鈥� says Tang. But, she notes:

鈥淥ne of the major issues with them is that we don鈥檛 know how they came up with these answers.鈥�

GOPHER helped Koo and his team dig up the parts of AI algorithms that drive reliability, performance, and accuracy. The findings help define the key building blocks for constructing the most efficient AI algorithms going forward. 鈥淲e hope this will help people in the future who are new to the field,鈥� says Shushan Toneyan, another graduate student at the Koo lab.

Imagine feeling unwell and being able to determine exactly what鈥檚 wrong at the push of a button. AI could someday turn this science-fiction trope into a feature of every doctor鈥檚 office. Similar to video-streaming algorithms that learn users鈥� preferences based on their viewing history, AI programs may identify unique features of our genome that lead to individualized medicine and treatments. The Koo team hopes GOPHER will help optimize such AI algorithms so that we can trust they鈥檙e learning the right things for the right reasons. Toneyan says:

鈥淚f the algorithm is making predictions for the wrong reasons, they鈥檙e not going to be helpful.鈥�

Written by: Luis Sandoval, Communications Specialist | [email protected] | 516-367-6826

Funding

Simons Center for Quantitative Biology at 黑料吃瓜资源, National Institutes of Health

Citation

Toneyan, S., Tang, Z., et al., 鈥淓valuating deep learning for predicting epigenomic profiles鈥�, Nature Machine Intelligence, December 5, 2022. DOI: