Extended Data Fig. 10: Refining on low Neff targets doesn’t change results substantially.
From: Protein structure prediction with in-cell photo-crosslinking mass spectrometry and deep learning
a- Performance improvements we observed are due to adding crosslinking information, not additional fine-tuning of the AlphaFold2 weights on low Neff targets. There are few outliers on both sides. The performance is virtually identical for targets without MSAs (TM-score average: AlphaLink = 0.322, AlphaFold2 = 0.308, Z-statistic = −0.072). b- Performance improvements we observed are due to adding crosslinking information, not additional fine-tuning of the AlphaFold2 weights on low Neff targets. There are few outliers on both sides. The performance is virtually identical for Neff = 10 (TM-score average: AlphaLink = 0.701, AlphaFold2 = 0.702, Z-statistic = 0.033). Points show the mean (N = 10) over 10 MSA subsamples.
