In the study of evolutionary biology and protein engineering, fitness landscapes are essential to understanding how genetic variations affect an organism’s survival and reproduction. Mapping these landscapes involves assessing the fitness associated with numerous genotypes, which can be challenging with traditional methods.
One researcher from the University of Zurich has turned to deep learning as a solution. Deep learning models, such as multilayer perceptrons, recurrent neural networks, and transformers, have been used to predict the fitness of genotypes based on experimental data. These models operate by training on a subset of genotypes with known fitness values and using this information to predict the fitness of a larger set.
The study revealed that deep learning models are highly effective, explaining over 90% of fitness variance in the data. This research represents a significant step forward in fitness landscape studies, offering a more scalable and efficient approach to mapping the complex relationship between genotypes and fitness. It also emphasizes the importance of sampling strategies in optimizing the performance of deep learning models.
This opens new avenues for evolutionary biology and protein engineering research, indicating a potential paradigm shift in how fitness landscapes can be studied and understood. These findings are significant for researchers and professionals in the field of AI and biology, as they offer a more effective way to analyze fitness landscapes and contribute to advancements in evolutionary biology and protein engineering.
For more information on this research, you can read the paper here.