A deep learning model for analyzing noisy biological systems

Abstract Biological systems are inherently noisy so that two genetically identical cells in the exact same environment will sometimes behave in dramatically different ways. This imposes a big challenge in building traditional supervised machine learning models that can only predict determined phenotypic variables or categories per specific input condition. Furthermore, biological noise has been proven to play a crucial role in gene regulation mechanisms. The prediction of the average value of a given phenotype is not always sufficient to fully characterize a given biological system. In this study, we develop a deep learning algorithm that can predict the conditional probability distribution of a phenotype of interest with a small number of observations per input condition. The key innovation of this study is that the deep neural network automatically generates the probability distributions based on only few (10 or less) noisy measurements for each input condition, with no prior knowledge or assumption of the probability distributions. This is extremely useful for exploring unknown biological systems with limited measurements for each input condition, which is linked not only to a better quantitative understanding of biological systems, but also to the design of new ones, as it is in the case of synthetic biology and cellular engineering.