An analytics system uses metagenomics to generate predictions indicating performance of biological or physical samples. In an embodiment, a method includes determining sequence data of a soil sample. The method further includes determining a plurality of features of the soil sample using the sequence data. The plurality of features is determined based at least in part on a measure of a first microbe detected in the soil sample and a different measure of a second microbe detected in the soil sample. The method further includes inputting the plurality of features to a model trained using measures of the first microbe and the second microbe detected in a plurality of soil samples. The method further includes generating, by the model using the plurality of features, a prediction of physical attribute of a plant grown in the soil sample.