23rd Human Proteome Organization World Congress

Integrating AI throughout a mass spectrometry-driven drug discovery approach

Keynote speaker

Keynote lecture

Professor Dr

Deep learning for peptide property prediction and beyond

Keynote Lecture

Various deep learning-assisted data analysis pipelines developed recently are able to boost the performance of mass spectrometry-based proteomics. One such pipeline is Oktoberfest that can assist the analysis of DDA, DIA, and PRM data. Oktoberfest is based on Prosit, a deep neural network that predicts multiple peptide properties. Since its release in 2019, it was used to generated &gt;20 billion predicted spectra. Here, we summarize the available Prosit models for peptide property predictions and specifically highlight recent developments to generate a new single model for fragment ion intensity prediction covering tryptic and non-tryptic, modified and unmodified, and labeled and unlabeled peptides for different mass analyzers and fragmentation methods.The application of deep learning in proteomics is still in its early stages. While it holds the potential to assist upstream and downstream analysis and interpretation of proteomics data even further, the lack of reproducibility and FAIRness of models are widely acknowledged concerns. To limit such issues, we have developed a number of re-usable open-source packages relevant to the development and application of deep learning models in proteomics. Beyond the development of novel peptide property predictors, we will showcase the potential of AI on use cases such as interpretable AI and drug response prediction.

Professor

End-to-end transfer learning in DIA proteomics for the characterization of unseen peptide modifications

New Technology: AI and Bioinformatics in Mass Spectrometry

Data independent acquisition (DIA) based proteomics has established itself as key for fast and deep characterization of complex proteomes. Identification of peptides and proteins from DIA data is commonly performed using deep-learning generated spectral libraries which were trained with a large corpus of known spectral matches. DIA search is therefore challenged when the properties of unseen post translation modifications or instrument setups should be predicted. Based on our recently published modular open-source search framework alphaDIA, we have introduced a new strategy for DIA search termed end-to-end transfer learning.AlphaDIA is an open-source framework for complete DIA search workflows. It builds on the scientific python stack and the alphaX ecosystem allowing flexible search strategies as well as default workflows accessible through a Python API, Jupyter notebooks, a command line interface. AlphaDIA encompasses the entire workflow, from raw file processing to protein quantity reporting, and supports files and proprietary formats from major vendors. Engineered for streamlined processing of large cohorts and diverse data sizes, it operates seamlessly on Windows, Linux, and Mac platforms, or can be deployed in a distributed manner in the cloud using Slurm or Docker.By integrating deep-learning prediction using our deep-learning framework AlphaPeptDeep and with DIA search, custom models are trained for specific experimental setups. Following an initial search, confident identifications are re-quantified and their fragment spectra are collected. Our robust and automated deep-learning pipeline optimizes the neural network for retention time, ion mobility, charge state and fragment spectra prediction while ensuring generalization through a held out test set. In a second search alphaDIA uses this trained model to increase identifications and decrease false discoveries.We show the potential of this technology by applying it to chemical as well as post translational modifications. In our experiments, end-to-end transfer learning can increase the number of identified precursors drastically by close to 50% percent, leading to more than 25% additional protein identifications. Rigorous benchmarking established that the effects aren&quot;t the result of overfitting but reflect generalization of the network to the new data distribution. To unlock the potential, we show the potential application to acetylated peptide discovery, HLA-peptide characterization and non-tryptic digests.

Oral presentation

Prof.

Causal inference enables the estimation of outcomes of interventions from observational mass spectrometry (MS)-based proteomics experiments

Understanding the proteome&quot;s response to perturbations is an important step towards a full understanding of its function. Traditionally this is done experimentally, by perturbing the system using external mechanisms such as drugs and observing the resulting effects on protein abundance. Alternatively, machine learning methods, such as neural networks can be used to estimate the effect of perturbations by considering joint coregulation of many proteins. These approaches can be successful in some circumstances, however, they do not incorporate prior knowledge about mechanisms between proteins. In contrast, causal inference methods make explicit use of prior knowledge and have been shown to perform well in the estimation of perturbations, for example, using transcriptomics data. These methods take as input abundances from an observational experiment (i.e., an experiment that does not implement a perturbation), and a graph of known causal relationships, and estimate the impact of perturbations on the graph. Here we describe a novel approach to translating causal inference methods to mass spectrometry (MS)-based proteomics, moving beyond the characterization of associations.The implementation of causal inference methods in the context of MS-based proteomics has a number of computational challenges that must be overcome in order for the methods to be effective. First, MS proteomics data is inherently noisy, is subject to batch effects and missing values, and quantifies peptide-fragments rather than proteins. Second, creating an accurate graph of causal relationships between proteins is non-trivial. We leverage the INDRA database of biological mechanisms which combines the content of pathway knowledgebases with literature mining to extract causal relations between the proteins quantified in the experiment. We employ a variety of filtering strategies to create a context-specific network with edges that support the data generating process. Finally, we develop a Bayesian model that leverages the network to model the relationships between proteins and is fit to the observational data, to perform model-based inference.We validate this method on synthetic data, then demonstrate it on real biological experiments. We highlight areas where machine learning methods such as neural networks fail to accurately predict the effect of interventions and show how causal inference can overcome this. We use computer simulations to investigate the circumstances in which the causal inference methods are likely to succeed, such as favorable network topology and number of biological replicates. Finally, we show the accuracy of the method using an experiment which investigates the effect of drug compounds on the chromatin-binding activity of transcription factors. In this setting, the causal model is trained on the observational DMSO control data and evaluated by comparing the model&quot;s predictions of the effect of drug interventions against experimental measurements.

Prof. Dr.

Prof.,Dr.

Privacy-preserving multi-center differential protein abundance analysis with FedProt

Expanding proteomics data availability unlocks potential for large-scale biomedical research, but integrating patient-derived mass spectrometry data can be problematic due to privacy concerns. Protein expression profiles, just like transcriptomics data, can be subject to genotype reconstruction attacks and must be treated as confidential information [1]. To enable privacy-preserving, robust analysis of distributed proteomic data despite data heterogeneity, we designed FedProt — the first tool for federated differential protein abundance analysis.FedProt represents the mathematical equivalent of DEqMS, a state-of-art modification of the limma method [2], and utilizes the hybrid approach of federated learning (FL) [3] and additive secret sharing [4]. In FL, multiple parties holding sensitive data jointly participate in the computation without revealing their data. This involves splitting the workflow into steps performed locally and aggregating results globally by a trusted server. In FedProt, this approach allows us to obtain the same result as centralized pooled data analysis, but without violating privacy.We evaluated FedProt using two multi-center datasets, a TMT human plasma dataset of 60 samples (2 conditions, 3 cohorts) and a LFQ dataset of 118 Escherichia coli samples (2 conditions, 5 cohorts). All MS data were uniformly preprocessed, quantified, and submitted to FedProt, which identified differentially abundant protein groups. We conducted a central DEqMS analysis as a baseline and compared the results of FedProt against it. Additionally, we studied class label and cohort size imbalance using simulated data.In all tests, FedProt effectively handled batch effects and produced results that closely matched the results of DEqMS baseline; fold-changes and negative log-transformed adjusted p-values were almost identical (maximal differences no greater than 1*10-11). It can analyze proteins absent in some cohorts and matches centralized analysis more precisely than typical meta-analyses like Fisher's or Stouffer's methods, RankProd, or the random effects model, representing the only privacy-preserving approach to differential protein abundance analysis to date.Using FL and additive secret sharing, FedProt effectively manages proteomic data complexity, including missing values and batch effects, enhancing sample sizes and statistical power while minimizing privacy risks without compromising accuracy. Accessible as a user-friendly FeatureCloud App (https://featurecloud.ai/app/fedprot), FedProt could catalyze larger collaborations and advance proteomics research, improving the robustness of differential protein abundance analysis results.Geyer, P.E., et al. Mol Cell Proteomics. 20, 2021, 100035.Zhu, Y., et al. Mol Cell Proteomics. 19, 2020, 1047–1057.McMahan, B., et al. PMLR, AI and Statistics, 2017, pp. 1273–1282.Cramer, R., et al., 2015. SMPC. Cambridge University Press.

Conference room 1-2

Session

AI and bioinformatics in mass spectrometry

Session
AI and bioinformatics in mass spectrometry

Termin

Vorsitzende

Programm

SessionAI and bioinformatics in mass spectrometry

Termin

Vorsitzende

Programm

Session
AI and bioinformatics in mass spectrometry