Supplementary Figure 2 shows figures related to the Consensus clustering analyses. (A) A delta area plot displays the relative change in the cumulative distribution function (CDF) curve comparing k and k-1 clusters from our cohort. (B) The silhouette plot displays the silhouette width for the four clusters determined by consensus clustering within this cohort. (C) Kaplan-Meier plot of the recurrence-free survival rates of the combined clusters (Clusters 1 with 3 and Clusters 2 with 4).

Supplementary Figure 6 shows a forest plot detailing the hazard ratio of the proteomic risk score and clinically relevant variables for PDA within our study cohort using multivariable Cox regression modeling.

Supplementary Figure 7 shows the proteomic signature performance based on the Receiver operator characteristic curve analysis. (A) Receiver operator characteristic curve (ROC) at 1 year of follow-up for the proteomic signature (purple) and other clinically relevant variables for PDA within our cohort. (B) An Area Under the Curve (AUC) plot displays the AUC as a function of time for the proteomic risk score (purple) and other clinically relevant variables for PDA within our cohort.

Supplementary Figure 9: Kaplan-Meier plot for patients dichotomized by a proteomic risk score that uses only the ten proteins detected in blood by mass spectrometry (PURB, GALM, SERPINA3, OAS3, KRT2, NUDT2, SERPINA4, CUTA, POSTN, CLEC11A).

Supplementary Figure 16 shows a Forest plot presentation of the hazard ratios for each KRT protein detected in the tumor samples from the study cohort, derived from a univariate Cox regression analysis with overall survival as the outcome of interest. Note that this figure excludes KRT75, which was detected in only two samples.

A schematic representing (A) data collection from 30-μm sections of 115 PDA and 61 adjacent normal fresh-frozen tissues that were prepared (19) for (B) DIA-MS. C, Protein quantification from DIA-MS data files using DIA-NN and MaxLFQ software for data normalization, QC, and peptide-to-protein inference and (D) the downstream analyses of the protein data. FC, fold change.

The distribution of clinical variables across the four clusters.

Supplementary Figure 12 shows a volcano plot displaying the differentially abundant proteins between tumors harboring a KRAS-G12D mutation versus those with any other KRAS-G12 mutation

Supplementary Table 9 shows the differential abundance of the KRT protein family across the different clinical variables of interest, where “Up” means upregulated and “Down” means downregulated in the group of interest versus others. *Note that KRT9, KRT10, KRT20, KRT23, KRT77, KRT74, KRT79 and KRT80 were not differentially abundant among any subgroup

Proteomic risk score for mortality. A, Forest plot detailing the multivariable HRs for the overall survival of each of the proteins used to build the proteomic risk score. B, Kaplan–Meier curve displaying the overall survival probability of the low- and high-risk groups within our cohort. C, Kaplan–Meier curve displaying the survival probability of the low- and high-risk groups within the CPTAC validation dataset.

Supplementary Figure 10 shows the differential abundance and pathway enrichment analyses based on KRAS mutations. (A) Volcano plot displaying the differentially abundant proteins between KRAS mutant PDA and KRAS wild-type PDA. (B) Pathways enriched in Gene Ontology molecular function database from upregulated proteins in tumors that harbored KRAS mutations.

Proteomic subtypes of PDA. A, A clustered heatmap displaying the highly variable protein intensities across the four clusters detected within this cohort. B, Kaplan–Meier plots of the survival rates of the four clusters. C, Kaplan–Meier plot of the survival rates of the clusters combined according to prognosis (cluster 1 with 3 and cluster 2 with 4).

Supplementary Table 4 shows previous evidence regarding each protein included in the risk score in terms of their association with different types of cancer diagnosis and prognosis. *Based on data extracted from the Human Protein Atlas (https://www.proteinatlas.org/)

List of the different differentially abundant proteins for the different groups of interest

Supplementary Figure 3 shows differential abundance and pathway enrichment analyses of clusters. (A) Volcano plot displaying the differentially expressed proteins between PDA samples classified between groups 1 and 3 against groups 2 and 4. (B) Pathways enriched in KEGG, Reactome, and Wikipathways for proteins upregulated in samples from clusters 2 and 4 compared to clusters 1 and 3.

Supplementary Table 2 shows the distribution of the clinical variables across the four proteomic-based clusters. Gln61Arg: Glutamine to Arginine at position 61, Gln61His: Glutamine to Histidine at position 61, Gly12Ala: Glycine to Alanine at position 12, Gly12Arg: Glycine to Arginine at position 12, Gly12Asp: Glycine to Aspartic Acid at position 12, Gly12Cys: Glycine to Cysteine at position 12, Gly12Leu: Glycine to Leucine at position 12, Gly12Val: Glycine to Valine at position 12. Note that all patients showed positivity for COSMIC signature 1, while no patients showed positivity for COSMIC signatures 6, 20, 25, and 26. For COSMIC signatures 13, 18, 28, and 30, only one patient was in the positive group.

DAPs and dysregulated pathways in PDA tumors compared with adjacent normal tissue. A, DAPs (n = 395; P < 0.05) between PDA and adjacent normal tissue. B, Top enriched pathway proteins upregulated in PDA tissue. KEGG, Kyoto Encyclopedia of Genes and Genomes. C, Gene linkage network of enriched pathways from Gene Ontology biological processes database for upregulated proteins in PDA tumors. FC, fold change.

Supplementary Figure 5 shows the proteomic information for the 18 risk-score proteins. (A) Peptide detection rate within the dataset. (B) Protein Missingness in our tumor data only. (C) Protein Missingness in the CPTAC Pancreatic Cancer Cohort.

Supplementary Figure 4 shows the Kaplan-Meier survival curves for each of the 18 proteins within the risk score. Median cut-off was used to dichotomize patients into two groups.

Supplementary Figure 8 shows Kaplan-Meier curves displaying (A) the three-year survival for groups dichotomised by proteomic risk score in ProCan data. (B) The recurrence-free survival for groups dichotomized by proteomic risk score in ProCan data.