Fleming Kretschmer

The paper evaluates coverage bias in small molecule datasets and proposes methods to assess and improve dataset representativeness for machine learning models.

The paper introduces a novel method that achieves 97.6% correct annotation of small molecule tandem mass spectra by combining spectral data with structural database information.

BERTax uses deep learning to classify DNA sequences taxonomically without requiring known database relatives, outperforming existing methods for novel organisms and combining with databases to improve accuracy.