Koehler, N. D. | bioRxiv Science

bioRxiv2020

Deep learning the collisional cross sections of the peptide universe from a million training samples

The size and shape of peptide ions in the gas phase are an under-explored dimension for mass spectrometry-based proteomics. To explore the nature and utility of the entire peptide collisional cross section (CCS) space, we measure more than a million data points from whole-proteome digests of five organisms with trapped ion mobility spectrometry (TIMS) and parallel accumulation - serial fragmentation (PASEF). The scale and precision (CV <1%) of our data is sufficient to train a deep recurrent neural network that accurately predicts CCS values solely based on the peptide sequence. Cross section predictions for the synthetic ProteomeTools library validate the model within a 1.3% median relative error (R > 0.99). Hydrophobicity, position of prolines and histidines are main determinants of the cross sections in addition to sequence-specific interactions. CCS values can now be predicted for any peptide and organism, forming a basis for advanced proteomics workflows that make full use of the additional information.

systems biology↗

bioRxiv2019

Self-supervised retinal thickness prediction enables deep learning from unlabeled data to boost classification of diabetic retinopathy

Access to large, annotated samples represents a considerable challenge for training accurate deep-learning models in medical imaging. While current leading-edge transfer learning from pre-trained models can help with cases lacking data, it limits design choices, and generally results in the use of unnecessarily large models. We propose a novel, self-supervised training scheme for obtaining high-quality, pre-trained networks from unlabeled, cross-modal medical imaging data, which will allow for creating accurate and efficient models. We demonstrate this by accurately predicting optical coherence tomography (OCT)-based retinal thickness measurements from simple infrared (IR) fundus images. Subsequently, learned representations outperformed advanced classifiers on a separate diabetic retinopathy classification task in a scenario of scarce training data. Our cross-modal, three-staged scheme effectively replaced 26,343 diabetic retinopathy annotations with 1,009 semantic segmentations on OCT and reached the same classification accuracy using only 25% of fundus images, without any drawbacks, since OCT is not required for predictions. We expect this concept will also apply to other multimodal clinical data-imaging, health records, and genomics data, and be applicable to corresponding sample-starved learning problems.

bioinformatics↗