Fig. 2

Prediction of recipient post-FMT microbiome properties from donor samples. A, B Different models evaluations scores of recipients’ Shannon, R2 scores (A), and SCC (B) over the human-to-GF cohorts (for parallel results on the different orders see Supplementary Material Fig. S7). The x-axis represents the model. The simplest models Ridge, KNN, and SVR are in blue; the networks and trees, RF, XGBOOST, and NN are in blue; the structure-based CNNs,  iMic1 and iMic2, are in pink. The standard errors over the 10 CVs are in black. iMic2 outperforms all the other models and predicts the recipient Shannon diversity with R2 of 0.358 and SCC of 0.6. C Scatter plot of the recipients’ predicted Shannon after FMT vs the real recipients’ Shannon over the human-to-GF cohorts. The black line represents the \(y=x\) line of a perfect match between the recipients’ predicted and real properties. D All predicted-real recipients’ properties SCCs over the human-to-GF cohorts, where the raw correlation between the recipient’s property and the donor’s property is in light blue and iMic’s improvement is in pink. E, F Different models evaluations scores of recipient’s Shannon diversity, R2 scores (E), and SCC (F) over the human-to-human cohorts (for parallel results on the different orders see Supplementary Material Fig. S9). The x-axis represents the model. The simplest models Ridge, KNN, and SVR are in orange; the networks and trees, XGBOOST, RF, and NN, are in orange; and the structure-based iMic2 is in pink. The standard errors over the 10CVs are in black. iMic2 outperforms all the other models and predicts the recipient Shannon diversity with R2 of 0.369 and SCC of 0.656. G Scatter plot of the recipients’ predicted Shannon post-FMT vs the real recipients’ Shannon diversity over the human-to-human cohorts. The black line represents the \(y=x\) line of a perfect match between the recipients’ predicted and real properties. H All predicted-real recipients’ properties SCCs over the human-to-human cohorts, where the raw correlation between the recipient’s property and the donor’s property is in orange and iMic’s improvement is in pink. I, J Histograms of AUCs of presence absence predictions of the species over the human-to-human cohorts (I) and of SCCs of the compositions predictions over the human-to-human cohorts (J). The x-axes represent the bins of scores, AUC (I)and SCCs (J), and the y-axes represent the number of different taxa that got that score. For parallel results on the human-to-GF cohorts see Supplementary Material Fig. S6K AUCs and SCCs of the 100 most frequent species. The left x-axis represents the presence or absence AUC over the human-to-human cohorts (pink), while the right x-axis represents the SCC (black). For parallel results on the human-to-GF cohorts, see Supplementary Material Fig. S6. L The prediction remains accurate long after the transplant in the human-to-human cohorts. The x-axis represents the SCC, and the y-axis represents the number of days post-FMT. There is not a significant difference in the SCCs over the time followed