The halls at a recent meeting of biologists in Seattle were buzzing with more than just the usual excitement about spiders, bats, bees, elephants and other creatures.
Researchers were also talking about the increased use of artificial intelligence and machine learning, at the the 2024 annual meeting of the Society for Integrative and Comparative Biology.
Such methods have life science applications beyond biomedical fields such as protein design, a more well-known use case. Researchers are leveraging AI to study how animals move their bodies, migrate, sense their environment, behave, and more.
“AI and machine learning methods are being used in diverse sub-disciplines in biology — from neuroscience, molecular biology, to animal behavior,” Jeff Riffell, a professor in the Biology Department at the University of Washington, told GeekWire.
Riffell and his colleagues presented an AI-powered system to study how insects detect odors in their environment. Their machine learning model predicts how moth neurons respond to different mixtures of smelly chemicals.
Shir Bar, who studies the intersection of biology and computer vision at Tel Aviv University, told GeekWire that she’s seeing more studies using AI for animal detection, tracking and behavioral classification, as well as in biomechanics for pose estimation (detecting position using computer vision methods).
Bar spoke at the meeting about how scientists can leverage AI, noting that entering the arena and finding the right tools for the task can be daunting. We asked Bar to identify some of the more outstanding AI/ML studies at the meeting, held earlier this month.
Bumblebee cooling
When the weather gets hot, bees keep the colony cool by fanning their wings. Researchers at the University of Wisconsin study this behavior by labeling individual bumblebees and tracking them with an automated imaging system while exposing them to high temperatures that simulate a three-day heatwave. The scientists integrate the tracking of individual bees with deep learning-based identification of fanning behavior. They are now using the system to test how bees respond to heat under different nutrient conditions. The research may help scientists understand how bees respond to climate change.
Insect treadmills
Researchers at Imperial College London place insects on small treadmills to measure how they move. At the meeting they also presented a synthetic dataset on such movement using three-dimensional models of insects, generated by a gaming engine, said Bar. According to the presenters, insects inspire researchers developing six-legged walking robots. After all, many insects can walk on ceilings and walls keep on going even if they lose limbs.
“This is a really innovative way to tackle the lack of training data that’s so prevalent in our field, especially since they are building a general system that is meant to work on diverse species of insects,” said Bar of the presentation.
Zebra tracking
An open-source tool to help capture animal behavior in the wild was showcased at the meeting by researchers at the University of Stuttgart and Princeton University. Smarter-labelme labels data used to train machine learning models, reducing the need to manually annotate datasets on animal movement. The researchers used the tool to quantify the activity of zebras from drone footage over large swaths of the savannah.
Seeing green
Scientists routinely label cellular molecules using green fluorescent protein (GFP), a laboratory tool originally derived from a jellyfish. Different color variants can arise from mutations in GFP, but exactly how has been unclear. Researchers have now developed a neural network model to predict the intensity of fluorescence from the underlying mutations in GFP, using protein folding parameters and other inputs. The approach could lead to the development of improved ways to visualize cellular molecules. This study was undertaken at the University of Maryland and the Janelia Research Campus of the Howard Hughes Medical Institute.
