PLM-SAV: A Δ-Embedding Approach for Predicting Pathogenic Single Amino Acid Variants

We have performed deep proteomic profiling down to as few as 9 Panc-1 cells using sample fractionation, TMT multiplexing, and a carrier/reference strategy. Off line fractionation of the TMT-labeled sample pooled with TMT-labeled carrier Panc-1 whole cell proteome was achieved using alkaline reversed phase spin columns. The fractionation in conjunction with the carrier/reference (C/R) proteome allowed us to detect 47 414 unique peptides derived from 6261 proteins, which provided a sufficient coverage to search for single amino acid variants (SAAVs) related to cancer. This high sample coverage is essential in order to detect a significant number of SAAVs. In order to verify genuine SAAVs versus false SAAVs, we used the SAVControl pipeline and found a total of 79 SAAVs from the 9-cell Panc-1 sample and 174 SAAVs from the 5000-cell Panc-1 C/R proteome. The SAAVs as sorted into high confidence and low confidence SAAVs were checked manually. All the high confidence SAAVs were found to be genuine SAAVs, while half of the low confidence SAAVs were found to be false SAAVs mainly related to PTMs. We identified several cancer-related SAAVs including KRAS, which is an important oncoprotein in pancreatic cancer. In addition, we were able to detect sites involved in loss or gain of glycosylation due to the enhanced coverage available in these experiments where we can detect both sites of loss and gain of glycosylation.

Single amino acid variants (SAAVs) in protein sequences are often a direct result of single-nucleotide polymorphisms (SNPs). Certain germline SAAVs have shown biological relevance in different disease conditions but lack precise quantification in circulation, which could hinder functional investigations and progress in biomarker development. Here, we have developed a multiplexed liquid chromatography-selected reaction monitoring (LC-SRM) assay that monitors 5 wild-type and variant peptide pairs (Complement Factor B: CFB-R32Q/R32W, Clusterin: CLU-N317H, Fetuin B: FETUB-K360R, and Kininogen: KNG1-L212P) in nondepleted human plasma. The assay was optimized for imprecision, linearity, stability, and calibration assessments with CVs of under 20%. The wild-type and variant peptide pairs were characterized in a set of healthy individual plasma samples. These target identifications were also validated by SNP genotyping with more than 99% accuracy. For all protein targets, we observed significantly lower concentrations of WT species in the presence variant peptides. In CFB, the concentration of R32Q was significantly lower than its counterpart R32W variant and WT species. Furthermore, our results distinguished phenotypes of homozygosity and heterozygosity of the SAAV presence through direct concentration level characterization. These findings provide some insights into how SAAVs affect quantitative assessments of target peptides. The assay demonstrates a platform for proteogenomic analyses with potential applications in both research and clinical settings.

Non-synonymous Single Nucleotide Variants (nsSNVs), resulting in single amino acid variants (SAVs), are important drivers of evolutionary adaptation across the tree of life. Humans carry on average over 10,000 SAVs per individual genome, many of which likely have little to no impact on the function of the protein they affect. Experimental evidence for protein function changes as a result of SAVs remain sparse – a situation that can be somewhat alleviated by predicting their impact using computational methods. Here, we used SNAP to examine both observed and in silico generated human variation in a set of 1,265 proteins that are consistently found across a number of diverse species. The number of SAVs that are predicted to have any functional effect on these proteins is smaller than expected, suggesting sequence/function optimization over evolutionary timescales. Additionally, we find that only a few of the yet-unobserved SAVs could drastically change the function of these proteins, while nearly a quarter would have only a mild functional effect. We observed that variants common in the human population localized to less conserved protein positions and carried mild to moderate functional effects more frequently than rare variants. As expected, rare variants carried severe effects more frequently than common variants. In line with current assumptions, we demonstrated that the change of the human reference sequence amino acid to the reference of another species (a cross-species variant) is unlikely to significantly impact protein function. However, we also observed that many cross-species variants may be weakly non-neutral for the purposes of quick adaptation to environmental changes, but may not be identified as such by current state-of-the-art methodology.

The most popular tools for predicting pathogenicity of single amino acid variants (SAVs) were developed based on sequence-based techniques. SAVs may change protein structure and function. In the context of van der Waals force and disulfide bridge calculations, no method directly predicts the impact of mutations on the energies of the protein structure. Here, we combined machine learning methods and energy scores of protein structures calculated by Rosetta Energy Function 2015 to predict SAV pathogenicity. The accuracy level of our model (0.76) is higher than that of six prediction tools. Further analyses revealed that the differential reference energies, attractive energies, and solvation of polar atoms between wildtype and mutant side-chains played essential roles in distinguishing benign from pathogenic variants. These features indicated the physicochemical properties of amino acids, which were observed in 3D structures instead of sequences. We added 16 features to Rhapsody (the prediction tool we used for our data set) and consequently improved its performance. The results indicated that these energy scores were more appropriate and more detailed representations of the pathogenicity of SAVs.

Proteome-wide onco-proteogenomic somatic variant identification in ER-positive breast cancer

Small variations in the primary amino acid sequence of extracellular matrix proteins can have profound effects on the biomineralization of hard tissues. For example, a change in one amino acid within the amelogenin protein can lead to drastic changes in enamel phenotype, resulting in amelogenesis imperfecta, enamel that is defective and easily damaged. Despite the importance of these undesirable phenotypes, there is very little understanding of how single amino acid variation in amelogenins can lead to malformed enamel. Here, we aim to develop a thermodynamic understanding of how protein variants can affect steps of the biomineralization process. High-resolution, in situ atomic force microscopy (AFM) showed that altering one amino acid within the murine amelogenin sequence (natural variants T21 and P41T, and experimental variant P71T) resulted in an increase in the quantity of protein adsorbed onto hydroxyapatite (HAP) and the formation of multiple protein layers. Quantitative analysis of the equilibrium adsorbate amounts revealed that the protein variants had higher oligomer-oligomer binding energies. MMP20 enzyme degradation and HAP mineralization studies showed that the amino acid variants slowed the degradation of amelogenin by MMP20 and inhibited the growth and phase transformation of HAP. We propose that the protein variants cause malformed enamel because they bind excessively to HAP and disrupt the normal HAP growth and enzymatic degradation processes. The in situ methods applied to determine the energetics of molecular level processes are powerful tools toward understanding the mechanisms of biomineralization.

MotivationExhaustive experimental annotation of the effect of all known protein variants remains daunting and expensive, stressing the need for scalable effect predictions. We introduce VespaG, a blazingly fast missense amino acid variant effect predictor, leveraging protein language model (pLM) embeddings as input to a minimal deep learning model.ResultsTo overcome the sparsity of experimental training data, we created a dataset of 39 million single amino acid variants from the human proteome applying the multiple sequence alignment-based effect predictor GEMME as a pseudo standard-of-truth. This setup increases interpretability compared to the baseline pLM and is easily retrainable with novel or updated pLMs. Assessed against the ProteinGym benchmark (217 multiplex assays of variant effect—MAVE—with 2.5 million variants), VespaG achieved a mean Spearman correlation of 0.48 ± 0.02, matching top-performing methods evaluated on the same data. VespaG has the advantage of being orders of magnitude faster, predicting all mutational landscapes of all proteins in proteomes such as Homo sapiens or Drosophila melanogaster in under 30 min on a consumer laptop (12-core CPU, 16 GB RAM).Availability and implementationVespaG is available freely at https://github.com/jschlensok/vespag. The associated training data and predictions are available at https://doi.org/10.5281/zenodo.11085958.

Massively parallel or second-generation sequencing-based genomic studies continuously identify new genomic alterations that may lead to novel protein sequences, which are attractive candidates for disease biomarkers and therapeutic targets after proteomic validation. Integrative proteogenomic methods have been developed to use mass spectrometry (MS)-based proteomics data for such validation. These methods replace the reference sequence database in proteomic database searching with a customized protein database that incorporates sample- or disease-specific sequences derived from DNA or RNA sequencing, thus enabling the identification of novel protein sequences. Although useful, this spectrum-centric approach requires a full evaluation of all possible spectrum-peptide pairs, which is time-consuming, error-prone, and difficult to apply. Here, we present PepQuery, a peptide-centric approach that focuses on only novel DNA or protein sequences of interest. PepQuery allows quick and easy proteomic validation of genomic alterations without customized database construction. We demonstrated the sensitivity and specificity of the approach in validating completely novel proteins, novel splice junctions, and single amino acid variants using simulations and experimental data. Notably, enabling unrestricted modification searching in PepQuery reduced false positives by up to 95%. We implemented PepQuery as both web-based and stand-alone applications. The web version provides direct access to more than half a billion MS/MS spectra from the Clinical Proteomic Tumor Analysis Consortium (CPTAC) and other cancer proteomic studies. The stand-alone version supports batch analysis and user-provided MS/MS data. PepQuery will increase the usage of proteogenomics beyond the proteomics community and will broaden the application of proteogenomics in personalized medicine.

Myxovirus resistance proteins (MxA and MxB) are interferon-induced proteins that exert antiviral activity against a diverse range of RNA and DNA viruses. In primates, MxA has been shown to inhibit myxoviruses, bunyaviruses, and hepatitis B virus, whereas MxB restricts retroviruses and herpesviruses. As a result of their conflicts with viruses, both genes have been undergoing diversifying selection during primate evolution. Here, we investigate how MxB evolution in primates has affected its restriction of herpesviruses. In contrast to human MxB, we find that most primate orthologs, including the closely related chimpanzee MxB, do not inhibit HSV-1 replication. However, all primate MxB orthologs tested restrict human cytomegalovirus. Through the generation of human and chimpanzee MxB chimeras we show that a single residue, M83, is the key determinant of restriction of HSV-1 replication. Humans are the only primate species known to encode a methionine at this position, whereas most other primate species encode a lysine. Residue 83 is also the most polymorphic residue in MxB in human populations, with M83 being the most common variant. However, ~2.5% of human MxB alleles encode a threonine at this position, which does not restrict HSV-1. Thus, a single amino acid variant in MxB, which has recently risen to high frequency in humans, has endowed humans with HSV-1 antiviral activity.

Identification of mutant proteins in biological samples is one of the emerging areas of proteogenomics. Despite the fact that only a limited number of studies have been published up to now, it has the potential to recognize novel disease biomarkers that have unique structure and desirably high specificity. Such properties would identify mutant proteoforms related to diseases as optimal drug targets useful for future therapeutic strategies. While mass spectrometry has demonstrated its outstanding analytical power in proteomics, the most frequently applied bottom-up strategy is not suitable for the detection of mutant proteins if only databases with consensus sequences are searched. It is likely that many unassigned tandem mass spectra of tryptic peptides originate from single amino acid variants (SAAVs). To address this problem, a couple of protein databases have been constructed that include canonical and SAAV sequences, allowing for the observation of mutant proteoforms in mass spectral data for the first time. Since the resulting large search space may compromise the probability of identifications, a novel concept was proposed that included identification as well as verification strategies. Together with transcriptome based approaches, targeted proteomics appears to be a suitable method for the verification of initial identifications in databases and can also provide quantitative insights to expression profiles, which often reflect disease progression. Important applications in the field of mutant proteoform identification have already highlighted novel biomarkers in large-scale investigations.

Barrier-to-Autointegration Factor 1 (Banf1/BAF) is a critical component of the nuclear envelope and is involved in the maintenance of chromatin structure and genome stability. Banf1 is a small DNA binding protein that is conserved amongst multicellular eukaryotes. Banf1 functions as a dimer, and binds non-specifically to the phosphate backbone of DNA, compacting the DNA in a looping process. The loss of Banf1 results in loss of nuclear envelope integrity and aberrant chromatin organisation. Significantly, mutations in Banf1 are associated with the severe premature ageing syndrome, Néstor–Guillermo Progeria Syndrome. Previously, rare human variants of Banf1 have been identified, however the impact of these variants on Banf1 function has not been explored. Here, using in silico modelling, biophysical and cell-based approaches, we investigate the effect of rare human variants on Banf1 structure and function. We show that these variants do not significantly alter the secondary structure of Banf1, but several single amino acid variants in the N- and C-terminus of Banf1 impact upon the DNA binding ability of Banf1, without altering Banf1 localisation or nuclear integrity. The functional characterisation of these variants provides further insight into Banf1 structure and function and may aid future studies examining the potential impact of Banf1 function on nuclear structure and human health.

Herpesviruses present a major global disease burden. Understanding the host cell mechanisms that block viral infections, as well as how viruses can evolve to counteract these host defenses, is critically important for understanding viral disease pathogenesis. This study reveals that the major human variant of the antiviral protein myxovirus resistance protein B (MxB) inhibits the human pathogen herpes simplex virus (HSV-1), whereas a minor human variant and orthologous MxB genes from even closely related primates do not. Thus, in contrast to the many antagonistic virus-host interactions in which the virus is successful in thwarting the host's defense systems, here the human gene appears to be at least temporarily winning at this interface of the primate-herpesvirus evolutionary arms race. Our findings further show that a polymorphism at amino acid 83 in a small fraction of the human population is sufficient to abrogate MxB's ability to inhibit HSV-1, which could have important implications for human susceptibility to HSV-1 pathogenesis.

Using sensors to accurately identify the user’s activity is helpful to select suitable indoor localization algorithm, thus improving the positioning accuracy. Traditional user activity recognition technology generally needs to artificially analyze and extract the data features, and sometimes it is difficult to accurately excavate the characteristics of complex user’s activity. By making use of the ability of automatically extracting feature of raw data with deep learning algorithm, this paper proposes a novel algorithm for complex user activity recognition based on CNN. As a representative neural network of the deep learning, CNN provides an end-to-end learning model with high recognition accuracy. The algorithm utilizes the original data collected by sensors embedded in a smartphone, such as acceleration sensor, gyroscope sensor and magnetic sensor, as input of neural network after 2D image-like conversion. By training and optimizing the various super parameters of the convolution neural network, such as the layer number of the convolution layer and the number of filters, the features of different activities are automatically learned by CNN. At the same time, we select ReLU as activation function to improve the training speed, and finally build a multiple user complex activity recognition model based on convolution neural network. As a comparison, we implements an activity recognition algorithm based on AdaBoost by using the handcrafted features. Extensive experimental results show that our proposed algorithm can accurately identify six kinds of complex behavior modes, such as walking normallywalking sideways, walking backward, crawling on the floor, pushing a pushcart, going up and down the elevator, and the accuracy is up to 99.8%, which is superior to the AdaBoost based activity recognition algorithm

This paper presented a hybrid neural network model to integrate information entropy theory and ant colony clustering for load forecasting. As short-term load forecasting is a complex problem with multifactor in power system, if all these factors are used as inputs of neural network, it will not only result in complicated network structure, but also long learning time and inaccurate prediction. First, information entropy theory is used to select relevant ones from all load influential factors, the results are used as inputs of neural network. It can reduce irrelevant load influential factors and the input variables of the input layer for neural network. Next, considering the features of power load and reduced influential factors, using ant colony clustering method, the practical historical load data within one year is divided into several groups. A separate module based on neural networks models each group. Then, the typical samples in each clustered group were selected as the training set for the separate improved Elman neural network which is a kind of globally feed forward locally recurrent network model with distinguished dynamical characteristics. According to the procedures, the reduced input variables and the typical training samples for each neural network can be gotten. Thus the neural network forecasting model based on information entropy and ant colony clustering can be constructed which can effectively reduce the training time and improve convergent speed. During the forecasting process, pattern recognizing is employed to activate the corresponding module for hourly load forecasting. The presented model was tested using Hebei Province daily load data, and the satisfactory results were obtained.

Epitope fine specificity of human anti-HLA-A2 antibodies identification of four epitopes including a haptenlike epitope on HLA-A2 at lysine 127