Inferring microbial communities using constrained damped lasso regression based on the generalized Lotka-Volterra model

In this study, we developed a new method to infer microbial communities. Synthetic and natural microbialcommunities play essential roles in the industry and our health. We used the generalized Lotka-Volterra(GLV) model to model the interaction between microbes. Due to the meaningfulness of the parameters inthis model, it has been widely used for modeling microbial communities. To the best of our knowledge, mostof the regression-based methods on the GLV equation did not consider the sparsity and constraints of the realproblem. Hence, to solve the limitations of the available methods, we developed damped lasso regularizationto solve this constrained-based convex optimization problem. We used CVX solver for this problem. Wetrained and tested our method on various simulated microbial communities with ۳ to ۵ interacting microbeswith different dynamics, including stable fixed point, limit cycle, and chaotic dynamics. We used the crossvalidationmethod to test our method's performance in inferring the magnitude and sign of the interactions.We calculated the correlation of the estimated abundance of interacting microbes based on the inferred modelwith actual data. Our results demonstrated that the developed method could accurately predict the parameterssign and magnitude. Furthermore, the correlation between the estimated abundance and real abundance wasmore than ۰.۹. We also evaluated the model performance in presence of noise.

کلیدواژه ها:

microbial community ، inference ، lasso regression ، GLV ، Optimization