Machine Learning Integration in Biological Circuit Development by Biomedical Engineers
Biomedical engineers at Duke University have developed a novel approach to utilize machine learning in modeling the intricate interactions occurring between complex variables in engineered bacteria. This groundbreaking method offers the potential to be applied across various biological systems, simplifying a modeling process that was previously too complex.
The researchers' breakthrough was published in the journal Nature Communications on September 25. The team focused on a biological circuit integrated into a bacterial culture to predict circular patterns. This new approach proved to be significantly faster than traditional methods, with runtimes 30,000 times quicker than the current computational model.
To enhance accuracy, the machine learning model was retrained numerous times. The researchers then compared the outputs from their model and tested it on a second, computationally distinct biological system. This confirmed the model's versatility outside of the initial problem set.
Professor Lingchong You, a biomedical engineer at Duke, explained the inspiration behind this research: "This work was inspired by Google showing that neural networks could learn to beat a human in the board game Go." She continued, "Even though the game has simple rules, there are far too many possibilities for a computer to calculate the best next option deterministically. I wondered if such an approach could be useful in coping with certain aspects of biological complexity."
The study incorporated 13 bacterial variables, including growth rates, diffusion, protein degradation, and cellular movement. A single computer would require at least 600 years to calculate six values per parameter with traditional methods. In contrast, the new machine learning system can complete the calculation in hours.
Postdoctoral associate Shangying Wang applied a deep neural network capable of making predictions much quicker than the original model. The network utilized model variables as input, assigning random weights and biases to make a prediction about the bacterial colony's pattern. With retraining from new data, the network fine-tuned its predictions, achieving accuracy with each iteration.
Upon finding similar predictions from four trained neural networks, the researchers determined they did not necessarily need to validate each answer using the slower computational model. Instead, they employed a "wisdom of the crowd" approach, relying on the collective responses from their neural networks.
After sufficient training, the biomedical researchers applied the machine learning model to a biological circuit, using 100,000 data simulations to train the neural network. The model identified one result featuring a bacterial colony with three rings. Furthermore, it pinpointed crucial variables necessary for this formation.
"The neural net was able to find patterns and interactions between the variables that would have been otherwise impossible to uncover," said Wang.
Finally, the researchers tested the system on a random biological system, which traditionally required a computer model that repeated specific parameters until it identified the most probable outcome. The new system demonstrated similar capabilities, suggesting its applicability to diverse complex biological systems.
The Duke team is now focusing their efforts on more intricate biological systems, aiming to further optimize the algorithm for increased efficiency. Postdoctoral associate Shangying Wang noted, "Our first goal was a relatively simple system. Now we want to improve these neural network systems to provide a window into the underlying dynamics of more complex biological circuits."
The machine learning model developed by Biomedical engineers at Duke University, as published in Nature Communications, has the potential to be applied across various biological systems, not just engineered bacteria. This model, trained with artificial intelligence, can simplify the modeling process and provide insights into complex medical-conditions and health-and-wellness, given its ability to predict patterns and interactions between variables quickly. Moreover, due to its versatility, demonstrated by testing the model on multiple systems, this technology could revolutionize the study of science, particularly in the field of medicine and health research.
