Artificial intelligence continues to make inroads into many aspects of our lives. But what about biology, the study of life itself? AI can examine hundreds of thousands of genomic data points to identify potential new therapeutic targets. Although these genomic insights can be helpful, scientists are not sure how today’s AI models reach their conclusions in the first place. Now a new system called SQUID arrives on the scene, armed to open the black box of AI’s dark internal logic.
SQUID (Surrogate Quantitative Interpretability for Deepnets) is a computational tool created by scientists at Cold Spring Harbor Laboratory (CSHL). It is designed to help interpret how AI models analyze the genome. Compared to other analysis tools, SQUID is more consistent, reduces background noise, and can lead to more accurate predictions about the effects of genetic mutations.
How can I make it work even better? CSHL Assistant Professor Peter Koo says SQUID’s professional training is key.
“The tools that people use to understand these models primarily come from other fields, like computer vision or natural language processing. They can be useful, but they’re not optimal for genomics. What we’ve done with SQUID has been decades of use. Quantitative “Knowledge of genetics helps us understand what these deep neural networks are learning,” explains Koo.
SQUID works by first creating a library of over 100,000 variant DNA sequences. They then use a program called MAVE-NN (Multiplex Assays of Variant Effects Neural Network) to analyze the library of mutations and their effects. This tool allows scientists to conduct thousands of virtual experiments simultaneously. In fact, they can “discover” the algorithm behind the most accurate predictions of a given AI. Their computational “catch” could set the stage for experiments that are more grounded in reality.
“In silico (virtual) experiments cannot replace real laboratory experiments. They can nevertheless be very informative, helping scientists hypothesize about how specific regions of the genome work or whether mutations may have clinically relevant effects. It can help build .” CSHL Associate Professor Justin Kinney, co-author of the study, explains:
There are countless AI models in the ocean. More people are entering the water every day. Koo, Kinney and their colleagues hope that SQUID will help scientists figure out what best meets their professional needs.
Despite being mapped, the human genome remains an incredibly difficult area. SQUIDs can help biologists explore the field more effectively and get closer to the true medical implications of their findings.
