PHILADELPHIA - A team of bioengineers from the University of Pennsylvania Institute for Medicine and Engineering have trained a computer neural network model to accurately predict how blood platelets would respond to complex conditions found during a heart attack or stroke.
Using an automated, robotic system, they exposed human blood platelets to hundreds of different combinations of biological stimuli like those experienced during a heart attack. This was done by fingerprinting each platelet sample with 34,000 data points obtained in response to all possible pairs of stimuli.
The team applied the system to predict intracellular calcium signaling responses of human platelets to any combination of up to six different agonists used at different dosages and even applied at different times. The model predicted platelet responses accurately, even distinguishing between 10 blood donors, demonstrating an efficient approach for predicting complex chemical responses in a patient-specific disease milieu.
The strategy involves selecting molecules that react with blood platelets under high-risk situations, such as a heart attack, measuring the cellular responses to all pairwise combinations of stimuli in a high-throughput manner and then training a two-layer, nonlinear, neural network with the measured cellular responses. For platelets, it was discovered that the complexity of integrating numerous signals can be built up from the responses to simpler conditions involving only two stimuli.
"With patient-specific computer models, it is now possible to predict how an individual's platelets would respond to thousands of 'in silico' heart-attack scenarios," said Scott L. Diamond, professor of chemical and biomolecular engineering and the director of the Penn Center for Molecular Discovery. "With this information we can identify patients at risk of thrombosis or improve upon current forms of anti-platelet therapies."
|Contact: Jordan Reese|
University of Pennsylvania