CoChat
Sign Up
Cited by 286Alzheimer's disease research and treatmentsAdvanced Proteomics Techniques and ApplicationsDementia and Cognitive Impairment Research

The DOI matches its Crossref record; 77 of the 78 references it cites were found in a registry or an index.

Proteome profiling in cerebrospinal fluid reveals novel biomarkers of Alzheimer's disease

Open access
Bader, Jakob M, Geyer, Philipp E, Müller, Johannes B, Strauss, Maximilian T, Koch, Manja, Leypoldt, Frank, Koertvelyessy, Peter, Bittner, Daniel, Schipke, Carola G, Incesoy, Enise I, Peters, Oliver, Deigendesch, Nikolaus, Simons, Mikael, Jensen, Majken K, Zetterberg, Henrik, Mann, Matthias·2020·Molecular Systems Biology
DOI 10.15252/msb.20199356PMC PMC7266499
Open this paper in CoChat — free accountOpen PDF

Abstract

Neurodegenerative diseases are a growing burden, and there is an urgent need for better biomarkers for diagnosis, prognosis, and treatment efficacy. Structural and functional brain alterations are reflected in the protein composition of cerebrospinal fluid (CSF). Alzheimer's disease (AD) patients have higher CSF levels of tau, but we lack knowledge of systems‐wide changes of CSF protein levels that accompany AD. Here, we present a highly reproducible mass spectrometry (MS)‐based proteomics workflow for the in‐depth analysis of CSF from minimal sample amounts. From three independent studies (197 individuals), we characterize differences in proteins by AD status (> 1,000 proteins, CV < 20%). Proteins with previous links to neurodegeneration such as tau, SOD1, and PARK7 differed most strongly by AD status, providing strong positive controls for our approach. CSF proteome changes in Alzheimer's disease prove to be widespread and often correlated with tau concentrations. Our unbiased screen also reveals a consistent glycolytic signature across our cohorts and a recent study. Machine learning suggests clinical utility of this proteomic signature.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.15252/msb.20199356

Cites

77 of 77 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Study overview and CSF proteome characterization Overview of the study populations (cohorts) and schematic proteomic workflow. The CSF of three cohorts comprising AD and control subjects was analyzed. The total number of subjects per cohort group is depicted. Light and dark shades represent female and male subjects, respectively. “Ctrl” refers to non‐AD control subjects.Number of proteins identified and quantified passing the 1% FDR cutoffs in each sample. Horizontal lines show the mean and the error bars ± SD. The dashed line indicates the level of the meta‐median (1,233 proteins) of the group medians of quantified proteins. Number of samples per group as shown in A).Data completeness curve. The number of proteins in the dataset (Y axis) depending on the minimum number of samples in which the proteins have each been quantified (X axis) is plotted. The arrows indicate 50%, 75%, and 100% data completeness.Median CSF protein abundance distribution as calculated from MS intensities of quantified peptides of each protein. The top ten most abundant proteins and hemoglobins are highlighted.Global correlation map of proteins generated by clustering the Pearson correlation coefficients of all possible protein combinations. The abundance of proteins with common regulation correlates across samples, and they therefore form a cluster. Prominent clusters are annotated with functional terms obtained from bioinformatics enrichment analysis. The position of tau (gene name MAPT) is labeled on the Y axis. The inset shows the color code for Pearson correlation coefficients.
Figure 1 Study overview and CSF proteome characterization Overview of the study populations (cohorts) and schematic proteomic workflow. The CSF of three cohorts comprising AD and control subjects was analyzed. The total number of subjects per cohort group is depicted. Light and dark shades represent female and male subjects, respectively. “Ctrl” refers to non‐AD control subjects.Number of proteins identified and quantified passing the 1% FDR cutoffs in each sample. Horizontal lines show the mean and the error bars ± SD. The dashed line indicates the level of the meta‐median (1,233 proteins) of the group medians of quantified proteins. Number of samples per group as shown in A).Data completeness curve. The number of proteins in the dataset (Y axis) depending on the minimum number of samples in which the proteins have each been quantified (X axis) is plotted. The arrows indicate 50%, 75%, and 100% data completeness.Median CSF protein abundance distribution as calculated from MS intensities of quantified peptides of each protein. The top ten most abundant proteins and hemoglobins are highlighted.Global correlation map of proteins generated by clustering the Pearson correlation coefficients of all possible protein combinations. The abundance of proteins with common regulation correlates across samples, and they therefore form a cluster. Prominent clusters are annotated with functional terms obtained from bioinformatics enrichment analysis. The position of tau (gene name MAPT) is labeled on the Y axis. The inset shows the color code for Pearson correlation coefficients.CC BY · PubMed Central
Figure EV1 Clinical information on cohort subjects AAge of participants at the time point of CSF collection. Bars represent mean and standard deviation. Participant numbers are 29, 31, 26, 12, 16, 33, 26, and 24 for Sweden AD, Sweden biochemical controls, Magdeburg AD, Magdeburg biochemical controls, Berlin AD, Berlin depression controls, and Berlin subjective cognitive impairment (SCI) controls.B–FCSF concentration of t‐tau (B), p‐tau181 (C), Aβ1–42 (D), Aβ1–40 (E), and the Aβ1–42/Aβ1–40 concentration ratio (F) as measured by ELISA. Bars represent mean and standard deviation. NA indicates that these data were not available. Participant numbers as in A) if data were available.G–ICSF concentrations of t‐tau plotted versus Aβ1–42 for samples of the Sweden (G), Magdeburg (H), and Berlin (I) cohorts. Samples classified as AD according to the biochemical criteria of this study colored in red, samples as non‐AD in blue.J–KCSF concentration of t‐tau plotted versus the Aβ1–42/Aβ1–40 concentration ratio for samples of the Magdeburg (J) and Berlin (K) cohorts. Samples classified as AD according to the biochemical criteria of this study colored in red, samples as non‐AD in blue.L Mini‐mental state examination (MMSE) scores, a measure of cognitive performance. Bars represent mean and standard deviation. These data were only available for the Berlin cohort. Participant numbers are 33, 26, and 24 for Berlin AD, Berlin depression controls, and Berlin subjective cognitive impairment (SCI) controls.
Figure EV1 Clinical information on cohort subjects AAge of participants at the time point of CSF collection. Bars represent mean and standard deviation. Participant numbers are 29, 31, 26, 12, 16, 33, 26, and 24 for Sweden AD, Sweden biochemical controls, Magdeburg AD, Magdeburg biochemical controls, Berlin AD, Berlin depression controls, and Berlin subjective cognitive impairment (SCI) controls.B–FCSF concentration of t‐tau (B), p‐tau181 (C), Aβ1–42 (D), Aβ1–40 (E), and the Aβ1–42/Aβ1–40 concentration ratio (F) as measured by ELISA. Bars represent mean and standard deviation. NA indicates that these data were not available. Participant numbers as in A) if data were available.G–ICSF concentrations of t‐tau plotted versus Aβ1–42 for samples of the Sweden (G), Magdeburg (H), and Berlin (I) cohorts. Samples classified as AD according to the biochemical criteria of this study colored in red, samples as non‐AD in blue.J–KCSF concentration of t‐tau plotted versus the Aβ1–42/Aβ1–40 concentration ratio for samples of the Magdeburg (J) and Berlin (K) cohorts. Samples classified as AD according to the biochemical criteria of this study colored in red, samples as non‐AD in blue.L Mini‐mental state examination (MMSE) scores, a measure of cognitive performance. Bars represent mean and standard deviation. These data were only available for the Berlin cohort. Participant numbers are 33, 26, and 24 for Berlin AD, Berlin depression controls, and Berlin subjective cognitive impairment (SCI) controls.CC BY · PubMed Central
Figure EV2 Robustness of the analytical workflow and enrichment analysis of the tau‐containing cluster in the global correlation map A–CComparison of inter‐participant variation and technical assay variation. Coefficients of variation (CVs) were determined in a separate experiment for intra‐plate CVs (A) and inter‐plate CVs (B) to benchmark protein quantitation. Biological CVs were calculated from the main study data (C). Proteins with a CV below 20% are highlighted in blue. Numbers of proteins above and below this CV cutoff are given above and below the cutoff line, respectively. The CV experiment data resulted from an independent protein search, and thus, the total number of identified proteins is not identical to the main study. The data show that technical variation is much smaller than inter‐participant variation.DAnnotation enrichment results for the tau (MAPT)‐associated cluster in Fig 1E. Enrichment in the cluster over the entire background CSF proteome vs. enrichment significance (−log10 of Benjamini–Hochberg‐adjusted P‐values). Terms of interest with links to neurons are highlighted in black.
Figure EV2 Robustness of the analytical workflow and enrichment analysis of the tau‐containing cluster in the global correlation map A–CComparison of inter‐participant variation and technical assay variation. Coefficients of variation (CVs) were determined in a separate experiment for intra‐plate CVs (A) and inter‐plate CVs (B) to benchmark protein quantitation. Biological CVs were calculated from the main study data (C). Proteins with a CV below 20% are highlighted in blue. Numbers of proteins above and below this CV cutoff are given above and below the cutoff line, respectively. The CV experiment data resulted from an independent protein search, and thus, the total number of identified proteins is not identical to the main study. The data show that technical variation is much smaller than inter‐participant variation.DAnnotation enrichment results for the tau (MAPT)‐associated cluster in Fig 1E. Enrichment in the cluster over the entire background CSF proteome vs. enrichment significance (−log10 of Benjamini–Hochberg‐adjusted P‐values). Terms of interest with links to neurons are highlighted in black.CC BY · PubMed Central
Figure 2 Differences in AD vs. non‐AD CSF proteome in the three cohorts A–CProtein AD/non‐AD fold changes plotted vs. statistical significance for Sweden (A), Magdeburg/Kiel (B), and Berlin (C) cohorts. Proteins associated with the GO annotation neuron projection labeled in orange. Proteins above the dashed green line are statistically significant (P < 0.05), and those above the black curves have a q‐value below 0.05 (see Materials and Methods).
Figure 2 Differences in AD vs. non‐AD CSF proteome in the three cohorts A–CProtein AD/non‐AD fold changes plotted vs. statistical significance for Sweden (A), Magdeburg/Kiel (B), and Berlin (C) cohorts. Proteins associated with the GO annotation neuron projection labeled in orange. Proteins above the dashed green line are statistically significant (P < 0.05), and those above the black curves have a q‐value below 0.05 (see Materials and Methods).CC BY · PubMed Central
Figure 3 CSF proteome alterations across the three cohorts A, BOn the left, number of proteins that differ significantly (P‐value < 0.05 in A; q‐value < 0.05 in B) in abundance by AD status within each cohort. On the right, number of proteins thereof that have a consistent directionality of either elevated or reduced abundance in AD CSF in pairwise comparisons of cohorts.C–ECorrelation of protein AD/non‐AD fold changes in pairwise combinations of two cohorts each. Combinations are Sweden vs. Magdeburg/Kiel (C), Sweden vs. Berlin (D), and Magdeburg/Kiel vs. Berlin (E). Proteins included differ significantly (P < 0.05) and consistently in abundance by AD status in both cohorts each.F, GProteins that differ significantly (P < 0.05 in E; q < 0.05 in F) in abundance by AD status across all three cohorts. Z‐scored abundances of proteins in the AD and non‐AD groups of all cohorts shown by the heat map (see Materials and Methods). Hierarchical clustering separates AD from non‐AD groups. Pyruvate kinase PKM (PKM) was quantified in two isoforms, and UniProt IDs are given in parentheses. Black frames highlight proteins with consistent AD/non‐AD fold changes across cohorts.
Figure 3 CSF proteome alterations across the three cohorts A, BOn the left, number of proteins that differ significantly (P‐value < 0.05 in A; q‐value < 0.05 in B) in abundance by AD status within each cohort. On the right, number of proteins thereof that have a consistent directionality of either elevated or reduced abundance in AD CSF in pairwise comparisons of cohorts.C–ECorrelation of protein AD/non‐AD fold changes in pairwise combinations of two cohorts each. Combinations are Sweden vs. Magdeburg/Kiel (C), Sweden vs. Berlin (D), and Magdeburg/Kiel vs. Berlin (E). Proteins included differ significantly (P < 0.05) and consistently in abundance by AD status in both cohorts each.F, GProteins that differ significantly (P < 0.05 in E; q < 0.05 in F) in abundance by AD status across all three cohorts. Z‐scored abundances of proteins in the AD and non‐AD groups of all cohorts shown by the heat map (see Materials and Methods). Hierarchical clustering separates AD from non‐AD groups. Pyruvate kinase PKM (PKM) was quantified in two isoforms, and UniProt IDs are given in parentheses. Black frames highlight proteins with consistent AD/non‐AD fold changes across cohorts.CC BY · PubMed Central
Figure EV3 Effect of control group subtypes, age, and gender on the proteomics results A, BAD versus non‐AD association of our 40‐protein signature does not depend on the different control groups within the Berlin (A) and Magdeburg/Kiel (B) cohorts. AD versus non‐AD fold changes for comparisons against the two distinct non‐AD control groups, i.e. AD/depression vs. AD/subjective cognitive impairment (all Berlin) and AD/Magdeburg ctrl vs. AD/Kiel ctrl, are plotted for the Berlin and Magdeburg/Kiel cohorts.CLinear regression model of protein intensity against AD/non‐AD status. Estimator strength for significant (P < 0.05) covariates age and sex shown by heat map. In case of age, the product of estimator (per 1 year) and the interquartile range (eleven years) in this dataset shown as strength. Overall explanation of variance by the regression model shown by R 2 values.
Figure EV3 Effect of control group subtypes, age, and gender on the proteomics results A, BAD versus non‐AD association of our 40‐protein signature does not depend on the different control groups within the Berlin (A) and Magdeburg/Kiel (B) cohorts. AD versus non‐AD fold changes for comparisons against the two distinct non‐AD control groups, i.e. AD/depression vs. AD/subjective cognitive impairment (all Berlin) and AD/Magdeburg ctrl vs. AD/Kiel ctrl, are plotted for the Berlin and Magdeburg/Kiel cohorts.CLinear regression model of protein intensity against AD/non‐AD status. Estimator strength for significant (P < 0.05) covariates age and sex shown by heat map. In case of age, the product of estimator (per 1 year) and the interquartile range (eleven years) in this dataset shown as strength. Overall explanation of variance by the regression model shown by R 2 values.CC BY · PubMed Central
Figure 4 Protein correlation with t‐tau measurements and analysis of annotation term enrichment A–CCorrelation of proteins with ELISA‐measured t‐tau concentration across samples within the Sweden (A), Magdeburg (B), and Berlin (C) cohorts. Proteins with a q‐value below 0.05 are labeled in yellow. Proteins of the 40‐protein signature are colored in red for those with higher abundance in AD CSF and in blue for those with higher abundance in non‐AD CSF.DThree‐cohort summary of proteins significantly correlating with ELISA‐measured t‐tau. Protein names given for the 29 proteins out of the 40‐protein signature with significant (P < 0.05) correlation in each of the three cohorts. Pyruvate kinase PKM (PKM) was quantified in two isoforms, and UniProt IDs are given in parentheses.EOverlap of proteins significantly differing by AD status with proteins significantly correlating to ELISA‐measured t‐tau.FAnnotation enrichment in the AD versus non‐AD fold change dimension. Terms with positive enrichment means are enriched in AD CSF over non‐AD CSF. Conversely, terms with enrichment means below zero are enriched in non‐AD compared with AD CSF. Annotations filtered for significance of enrichment (P < 0.05) and term size (10–100 proteins per term) in all three cohorts.G, HProtein abundance distribution of CSF (G) and brain (H) showing the abundances of AD‐modulated CSF proteins. Proteins of our 40‐protein signature are highlighted in red (elevated abundance in AD) and blue (elevated abundance in non‐AD). Proteins linked to glucose metabolism are highlighted in purple and labeled.
Figure 4 Protein correlation with t‐tau measurements and analysis of annotation term enrichment A–CCorrelation of proteins with ELISA‐measured t‐tau concentration across samples within the Sweden (A), Magdeburg (B), and Berlin (C) cohorts. Proteins with a q‐value below 0.05 are labeled in yellow. Proteins of the 40‐protein signature are colored in red for those with higher abundance in AD CSF and in blue for those with higher abundance in non‐AD CSF.DThree‐cohort summary of proteins significantly correlating with ELISA‐measured t‐tau. Protein names given for the 29 proteins out of the 40‐protein signature with significant (P < 0.05) correlation in each of the three cohorts. Pyruvate kinase PKM (PKM) was quantified in two isoforms, and UniProt IDs are given in parentheses.EOverlap of proteins significantly differing by AD status with proteins significantly correlating to ELISA‐measured t‐tau.FAnnotation enrichment in the AD versus non‐AD fold change dimension. Terms with positive enrichment means are enriched in AD CSF over non‐AD CSF. Conversely, terms with enrichment means below zero are enriched in non‐AD compared with AD CSF. Annotations filtered for significance of enrichment (P < 0.05) and term size (10–100 proteins per term) in all three cohorts.G, HProtein abundance distribution of CSF (G) and brain (H) showing the abundances of AD‐modulated CSF proteins. Proteins of our 40‐protein signature are highlighted in red (elevated abundance in AD) and blue (elevated abundance in non‐AD). Proteins linked to glucose metabolism are highlighted in purple and labeled.CC BY · PubMed Central
Figure EV4 MMSE score correlation analysis and proteome alterations when stratifying the Berlin cohort by MMSE score ACorrelation of proteins to the mini‐mental state examination (MMSE) scores in the Berlin cohort. Proteins with a correlation q‐value below 0.05 are labeled. Proteins of the 40‐protein signature are colored in red for proteins with increased abundance in AD CSF and in blue for proteins with increased abundance in non‐AD CSF.B–FCSF proteome alterations between groups of lower MMSE scores (poor neuropsychological performance) and groups of higher MMSE score as separated by cutoffs of 29 (A), 27 (B), 25 (C), 23 (D), and 21 (E). Proteins of the 40‐protein signature are colored in red for proteins with increased abundance in AD CSF and in blue for proteins with increased abundance in non‐AD CSF.
Figure EV4 MMSE score correlation analysis and proteome alterations when stratifying the Berlin cohort by MMSE score ACorrelation of proteins to the mini‐mental state examination (MMSE) scores in the Berlin cohort. Proteins with a correlation q‐value below 0.05 are labeled. Proteins of the 40‐protein signature are colored in red for proteins with increased abundance in AD CSF and in blue for proteins with increased abundance in non‐AD CSF.B–FCSF proteome alterations between groups of lower MMSE scores (poor neuropsychological performance) and groups of higher MMSE score as separated by cutoffs of 29 (A), 27 (B), 25 (C), 23 (D), and 21 (E). Proteins of the 40‐protein signature are colored in red for proteins with increased abundance in AD CSF and in blue for proteins with increased abundance in non‐AD CSF.CC BY · PubMed Central
Figure EV5 Comparison of the results of our study with the CSF1 cohort of Higginbotham et al Overlap of proteins identified as AD‐regulated in the two studies. Out of the 43 proteins significantly (P < 0.05 across 3 cohorts) associated with AD or non‐AD in this study, 40 exhibited consistent difference across 3 cohorts (40‐protein signature). Of these, 38 were present in the CSF1 dataset of the Higginbotham et al study containing 2,875 proteins. In that dataset 528 proteins were significantly (P < 0.05) different between AD and control, 26 of the 38 matched proteins found as potential AD markers in this study overlapped. All 26 proteins had a consistent association with either AD or non‐AD CSF across the CSF1 dataset and this study, 25 thereof were elevated in AD CSF and one protein elevated in non‐AD CSF.Contingency table for analysis of enrichment of our three‐cohort panel in the significantly AD vs. non‐AD altered proportion of the CSF1 dataset.Heat map comparing the AD versus non‐AD fold changes for the 26 protein intersection of significant (P < 0.05) proteins between our study and the CSF1 dataset. Proteins were ranked based on the mean fold change of our study.AD/non‐AD fold changes of significant (P < 0.05) and consistent proteins of this study correlate well with fold changes of the CSF1 cohort.
Figure EV5 Comparison of the results of our study with the CSF1 cohort of Higginbotham et al Overlap of proteins identified as AD‐regulated in the two studies. Out of the 43 proteins significantly (P < 0.05 across 3 cohorts) associated with AD or non‐AD in this study, 40 exhibited consistent difference across 3 cohorts (40‐protein signature). Of these, 38 were present in the CSF1 dataset of the Higginbotham et al study containing 2,875 proteins. In that dataset 528 proteins were significantly (P < 0.05) different between AD and control, 26 of the 38 matched proteins found as potential AD markers in this study overlapped. All 26 proteins had a consistent association with either AD or non‐AD CSF across the CSF1 dataset and this study, 25 thereof were elevated in AD CSF and one protein elevated in non‐AD CSF.Contingency table for analysis of enrichment of our three‐cohort panel in the significantly AD vs. non‐AD altered proportion of the CSF1 dataset.Heat map comparing the AD versus non‐AD fold changes for the 26 protein intersection of significant (P < 0.05) proteins between our study and the CSF1 dataset. Proteins were ranked based on the mean fold change of our study.AD/non‐AD fold changes of significant (P < 0.05) and consistent proteins of this study correlate well with fold changes of the CSF1 cohort.CC BY · PubMed Central
Figure 5 Machine learning separates AD from non‐AD CSF at high performance Decision tree to classify AD vs. non‐AD participants based on the protein levels of a core 26 protein set. Splits are indicated by black triangles. A tree with a minimum depth of six can correctly classify the participants by AD status.Receiver operating characteristic (ROC) curve for the model based on XGBoost. The diagonal line indicates random performance. Blue line represents the mean performance of the model when trained on six stratified train—test splits (k‐fold). The gray areas represent the standard deviation of ROC values.Confusion matrix indicating model performance when predicted on the test split of the cross‐validation. Overall accuracy is 0.85.
Figure 5 Machine learning separates AD from non‐AD CSF at high performance Decision tree to classify AD vs. non‐AD participants based on the protein levels of a core 26 protein set. Splits are indicated by black triangles. A tree with a minimum depth of six can correctly classify the participants by AD status.Receiver operating characteristic (ROC) curve for the model based on XGBoost. The diagonal line indicates random performance. Blue line represents the mean performance of the model when trained on six stratified train—test splits (k‐fold). The gray areas represent the standard deviation of ROC values.Confusion matrix indicating model performance when predicted on the test split of the cross‐validation. Overall accuracy is 0.85.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Cited by

Sorted by citations

See all 286

Related

See all

Metadata

Authors
Bader, J. M., Geyer, P. E., Müller, J. B., Strauss, M. T., Koch, M., Leypoldt, F., Koertvelyessy, P., Bittner, D., Schipke, C. G., Incesoy, E. I., Peters, O., Deigendesch, N., Simons, M., Jensen, M. K., Zetterberg, H., & Mann, M.
Affiliations
  • Max Planck Institute of Biochemistry (DE)
  • University of Copenhagen (DK)
  • Harvard University (US)
  • Christian-Albrechts-Universität zu Kiel (DE)
  • Otto-von-Guericke-Universität Magdeburg (DE)
  • Humboldt-Universität zu Berlin (DE)
  • German Center for Neurodegenerative Diseases (DE)
  • University Hospital of Basel (CH)
  • Sahlgrenska University Hospital (SE)
Corresponding author
Mann, Matthias
ORCID
Venue
Molecular Systems Biology
Publisher
Springer Science and Business Media LLC
Published
2020-06-01
Volume
16
Issue
6
Pages
e9356-e9356
ISSN
1744-4292 (ISSN-L)
DOI
10.15252/msb.20199356
PMC
PMC7266499
OpenAlex
W3030835098
Cited by
286
References
78
Citations by year
2026: 30 · 2025: 54 · 2024: 55 · 2023: 68 · 2022: 44 · 2021: 27 · 2020: 8
FWCI
22.41
Type
journal article
Language
en
Open access
Licence
Locations
  • Molecular Systems Biology (publishedVersion, other-oa)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Research at the University of Copenhagen (University of Copenhagen) (publishedVersion, cc-by)
  • MDC Repository (Max-Delbrueck-Center for Molecular Medicine) (acceptedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • UCL Discovery (University College London) (submittedVersion)
  • Open access LMU (Ludwid Maxmilian's Universitat Munchen) (submittedVersion, cc-by)
  • Multimedialen Archiv und Publikationsserver der Christian-Albrechts-Universität zu Kiel (Christian-Albrechts-Universität zu Kiel) (submittedVersion, cc-by)
  • mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich) (submittedVersion)
  • DZNE Pub (submittedVersion, cc-by)
  • MPG.PuRe (Max Planck Society) (submittedVersion, other-oa)
Funders
  • Novo Nordisk Fonden (NNF15CC0001, NNF15OC0016692)
  • EC|Horizon 2020 Framework Programme (825694)
  • Deutsche Forschungsgemeinschaft (412136960)
  • H2020 European Research Council (ERC‐2012‐SyG_318987)
  • Funding from the Harvard T.H. Chan School of Public Health Dean's Challenge Program sponsored by the McLennan Family Fund
Topics
  • Life Sciences › Neuroscience › Cellular and Molecular Neuroscience › Alzheimer's disease research and treatments
  • Physical Sciences › Chemistry › Spectroscopy › Advanced Proteomics Techniques and Applications
  • Health Sciences › Medicine › Psychiatry and Mental health › Dementia and Cognitive Impairment Research
Keywords
Alzheimer's disease, cerebrospinal fluid proteome, biomarkers, tau protein, SOD1, DJ-1, neurodegeneration, mass spectrometry-based proteomics, machine learning, clinical utility
MeSH terms
Machine Learning, Alzheimer Disease, Glycolysis, Humans, Nerve Degeneration, Neurons, Reproducibility of Results, Cohort Studies, Biomarkers, tau Proteins, Proteome, Proteomics

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review