Skip to main content
Fig. 4 | Journal of Cheminformatics

Fig. 4

From: MS2DeepScore: a novel deep learning similarity measure to compare tandem mass spectra

Fig. 4

We here define a high structural similarity as Tanimoto > 0.6 and explore how well high structural similarity pairs can be retrieved using various spectral similarities. Collecting all spectrum pairs from the test set (3601 spectra) with mass spectral similarity > X with X increasing from 0 to 1.0, we compute precision and recall for the different mass spectral similarity measures (MS2DeepScore, Spec2Vec, modified Cosine). The curves illustrate the tradeoff between higher recall (towards the right) and higher precision (towards the left). They also reveal that MS2DeepScore gives notably better precision/recall combination over the entire range, followed by Spec2Vec and only then modified Cosine

Back to article page