Proteomics 101 (cancer)
Under scientific review
This introductory page is under scientific review. Use it for orientation, not clinical decision-making.
Proteomics measures proteins and their modifications (e.g. phosphorylation, ubiquitination). It sits close to phenotype: what the cell is actually doing, not only what it is transcribing.
The Computational Challenge of Proteomics
Reading proteins is exponentially harder than reading DNA or RNA for a fundamental reason: There is no PCR for proteins. If a protein is in low abundance in the cell, we cannot "amplify" it for the sequencer to read. The equipment will only read the most abundant proteins (like albumin in the blood) unless you perform an enrichment (depletion) step.
Bottom-Up vs Top-Down
Bioinformatics handles two sequencing approaches via Mass Spectrometry (LC-MS/MS):
- Bottom-Up (Shotgun): The standard approach. Proteins are "chopped" by an enzyme (like Trypsin) into small pieces called peptides. The mass spectrometer reads these peptides. The computational challenge here is protein inference: trying to guess the original whole protein by assembling the peptide puzzle.
- Top-Down: Whole proteins are analyzed. Preserves incredible information about isoforms and PTMs, but the file size is massive and the software challenge of mapping whole-protein spectra remains a processing bottleneck.
The Data Flow (From m/z to Proteins)
Unlike Genomics which spits out letters (ATCG), a mass spectrometer spits out MS/MS Spectra: peak graphs representing the Mass-to-Charge (m/z) ratio of ions.
- The software (like MaxQuant or MSFragger) takes each spectrum (peak graph) from your experiment.
- It theoretically simulates on the computer what the spectra of all known human body proteins would look like (using the UniProt database).
- It calculates a similarity metric (Search Engine Score) between the real spectrum and the millions of simulated spectra to find a "Match".
Quantification: DDA vs DIA
When data engineers receive a "Protein Expression Matrix", the robustness of those numbers depends on how the machine was configured:
- DDA (Data-Dependent Acquisition): The machine is fast but random. If you run the same patient sample twice, the machine might read different peptides each time. It generates spreadsheets with many "Missing Values" (NaN), which is a nightmare for Machine Learning.
- DIA (Data-Independent Acquisition): The new gold standard (e.g., SWATH-MS). The machine reads everything in systematic slices. It produces much less noisy data, almost without "Missing Values", allowing ML models to compare patient cohorts accurately without needing severe data imputation techniques.
Post-Translational Modifications (PTMs)
The protein is not the end of the line. Cancer is driven by PTMs:
- Phosphoproteomics: Measures whether a protein is turned on/off via the addition of phosphorus. Essential for studying Kinase Inhibitor targets. The data matrix now doesn't map "Genes", but "Specific Sites" (e.g., BRAF_S602).
- Ubiquitination: The target of new Targeted Protein Degradation drugs (PROTACs).