Fig. 6

Optimal artifical mixture of grown microbes A GA schematic figure. The GA contains the following steps: Step A―Initial population. One hundred donor samples are randomly sampled from the donors of all the cohorts. Step B―Adding a binary vector to each parent donor. The binary vector consists of 1 when the ASV’s abundance is higher than 0, and is 0 otherwise. Step C―Evaluation of recipients’ FMT future result after a week. By applying the pre-trained iMic model to the parent donors, we get the future recipients’ outcomes. Step D―Selection. The selection is done according to our fitness function choosing the best 30 donors with the most appropriate recipients outcome. Step E―Reproduction. To complete the parents of the next generation a mutation occurs with a probability of 0.3, and recombination occurs with a probability of 0.3. Step F―Checking stopping rule. If the stopping criterion is met, the donors of step E are returned; otherwise, the new generation of donors from step E is again used for the outcome prediction using iMic’s in C, until the stopping criterion is met. B, C GA convergences within 25 epochs on the Shannon diversity optimization task for both maximizing (B) and minimizing the recipient’s Shannon diversity a week post-FMT (C). D Monitoring the number of non-zero taxa of donors during the maximizing optimization. The x-axis represents the number of non-zero taxa (log scale) and the y-axis represents the predicted Shannon diversity of the best donors. E SCCs between the property in the optimized donors and the predicted recipient. The significantly predicted orders from the validation experiment are in red. F, G Percentage of the most common taxa in the optimized donors for the Shannon diversity task for different \(\gamma\) values (F) and for different prediction tasks (G)