A proteome is the complete set of proteins produced in an organism, system or biological context; proteomics is the large-scale study of proteomes. The research divides into two families:

Two arguments — one from biology, one from economics:

The correlation is real but weak — measured values:

Proteins are built from 20 amino acids joined by peptide bonds into a chain; one or more chains fold into the 3D form. Structure determines function.

The lecture states it explicitly: the amino acid sequence is the foundation of protein structure, therefore most proteomics research focuses on primary-structure work — which is why the entire LC-MS/MS pipeline is built around determining peptide sequence first, then inferring the protein.

Single-protein structure prediction comes in three flavours: ab initio (energy-based; small proteins only), fold recognition (matches topology, so it works even at low sequence identity), homology modelling (template-based; degrades sharply below ≈30% sequence identity to the template).

Bottom-up (a.k.a. shotgun proteomics): digest proteins into peptides first, analyse the peptides, then infer the proteins. The overview slide's numbered path is the exam skeleton:

1. Extract protein from the sample — optional organelle enrichment or affinity purification
2. Digest proteins to peptides — optionally enrich for modifications ①
3. After LC separation, peptides are ionised ② and injected into a high-resolution mass spectrometer
4. MS1 spectra (peptide isotope patterns) are recorded on a cycle of roughly one second ③
5. Peptide precursors are selected, fragmented, and MS2 spectra recorded ④
6. MS1 and MS2 are written to disk — typically several GB per LC-MS run ⑤ — then analysed by proteomics software

Top-down analyses intact proteins without digestion, so it preserves complete proteoform information — at the cost of technical difficulty, limited dynamic range and shallower throughput.

The mixture is driven through a column packed with the stationary phase by the mobile phase under pressure. Because each component interacts with the stationary phase differently, components travel at different speeds and elute at different retention times. Separation is driven by differential affinity, not molecular size.

Practice: peptides are separated by reversed-phase (RP) chromatography — C18 stationary phase, water/acetonitrile gradient with formic or acetic acid. ESI is the dominant ionisation mode coupled to LC; MALDI is rarely attached to LC systems.

MS1 separates ions by m/z → a specific m/z is isolated → energised to fragment (MS2) → fragments are re-separated by m/z → the MS2 spectrum is recorded.

A PSM (peptide–spectrum match) is one measured MS2 spectrum matched to a theoretical spectrum from the database. Because the highest-scoring PSM can still be a false positive, workflows control FDR with the target–decoy approach:

1. Spectra are searched against the target database and a decoy database designed to produce false-positive matches
2. One standard way to build decoys: reverse the target sequences → peptides that do not occur in nature
3. Comparing the score distributions of target vs decoy PSMs yields posterior error probabilities, from which FDR is estimated and controlled

• LFQ (e.g. MaxLFQ) — infers relative protein abundance from peptide intensities
• Spectral counting — approximates abundance by the number of peptides identified; simple but depth-dependent
• PRM / SRM / MRM — targeted acquisition for high-precision quantification, and the route to absolute quantification
• 3D vs 4D proteomics — 3D separates by retention time, m/z and intensity; 4D adds ion mobility (CCS) to resolve peptides with similar m/z but different shape, size or charge

Once proteins are identified and quantified across many samples, the result is a matrix: rows = proteins (or protein groups), columns = samples, cells = abundance or abundance ratio.

Applications — the lecture's pipeline slide places proteomics across the whole development arc:

Flagship programme: CPTAC — the Clinical Proteomic Tumor Analysis Consortium (proteomics.cancer.gov/programs/cptac), a proteogenomic pipeline linking tissue source sites, proteome characterisation centres and data analysis centres.

Accessibility in practice — the kinase tree illustrates it: BCR-ABL in CML is a druggable target because imatinib fits its ATP pocket and blocks signalling. And subcellular localisation matters (reference: the Human Protein Atlas, proteinatlas.org): MYC sits in the nucleoplasm, while EGFR spans Golgi apparatus, plasma membrane, nucleus, cell junctions and more — same cell, very different accessibility.

① Small-molecule targets (Nature Reviews Drug Discovery, 2016): of small-molecule drugs by target family — GPCRs ≈33%, ion channels ≈18%, nuclear receptors ≈16%, kinases ≈3%, other ≈30%. GPCRs are the single richest target family — worth remembering as the opposite of the intuition that kinases dominate.

② ADC targets (ADCs entering clinical development in 2024): HER2 (9) leads, then B7H3 (5) and HER3 (5), then TROP2, Nectin4, 5T4 (4 each), DLL3 (3), and a long tail (PTK7, FRα, GPC3, B7H4, ROR1, NaPi2b, cMET, CD30, CD33, CDH17, CD228, CEA/CD66, CD56…). Bispecific combos also appear: TROP2×HER3, B7H3×PDL1, EGFR×TROP2, EGFR×cMET.

③ CAR-T target selection (J. Hematol. Oncol., 2019) — three determinants:

The CAR-T workflow (Cleveland Clinic figure): collect blood → separate T cells → genetically engineer them to express the chimeric antigen receptor → expand to millions of cells → (chemotherapy beforehand) → infuse back. Target illustration: CD19 and CD22 across B-cell development — the lineage markers behind B-cell malignancy therapy.

Protein amount is not one number from one layer; it is the output of stacked regulation:

Metabolites are the small-molecule substrates, intermediates and products of cell metabolism. “Small molecule” is the operative word: metabolites can be analysed by LC-MS directly, without digestion — the natural companion layer to proteomics.

• Blind men and the elephant — each researcher touches one part; any single omics is a partial truth, and only integration recovers the whole animal.
• “Can a biologist fix a radio?” (Cancer Cell, 2002) — a warning against pure reductionism: dismantling a system and studying components one at a time may discard the essential information (the wiring, not the parts).

A network is simply a set of interconnected nodes (node = a basic element; edge = a connection; mathematically G = {N, E}). Integrating multi-omics means hanging data from different layers onto one network and finding the nodes that several layers point at simultaneously. A hub is a highly connected node — in biology, usually functionally important and pleiotropic.

Analysis methods integrate layers by fusion: e.g. patient-similarity network fusion (Nature Methods, 2014) — build one similarity network per omics layer, then iterate to fuse them into a single patient network “supported by all data”.

The three correlation numbers on the same slide — they must be read together, never one at a time:

• Read the formula term by term — m = mass, v = velocity, z = charge, B = magnetic field strength. Hold v and B fixed and r scales with m/z, so the instrument is really an m/z ruler. This sits one line under everything else in the lecture: LC separates by chemistry (retention time), MS separates by m/z.
• MS/MS in the deck′s own words — the first stage is ordinary mass spectrometry; then the ions of one chosen m/z are collected again, energised so the molecules fragment, and the fragments are re-analysed to generate a secondary mass spectrum (MS2). Those fragments are the product ions; for peptides they are the b-ions (N-terminal prefix) and y-ions (C-terminal suffix).
• How the decoy is made — sequence reversing — the deck reverses the target peptide to give a sequence that does not occur in nature: target AAHTEDINACTLTTSPR → decoy PSTTLTCANIDETHAAR. Searching both databases and comparing the two score distributions yields posterior error probabilities, and that is what lets you set an FDR threshold (typically <1%).

Why trypsin specifically: cleaving after K and R gives peptides that (i) are in a convenient 6–20 residue length range for LC-MS, and (ii) end in a basic residue, so they carry a reliable positive charge and fragment predictably in MS2 — which is what makes the b-ion and y-ion ladders readable at all.

The trade-off: digestion destroys the connection between peptides. You infer which proteoform was present from which peptides you saw, and peptides shared between isoforms (or between protein family members) are inherently ambiguous — the well-known protein inference problem. Top-down proteomics, which measures intact proteins, avoids it but is far harder.

Why it belongs in a pharma course: acetylation is the mechanism behind a whole drug class — HDAC inhibitors keep histones acetylated and chromatin open, and BET inhibitors block the reader proteins that recognise acetylated lysines. It is also why the histone code is considered writable, not fixed.

The logic in one line: a decoy sequence is not present in the sample, so any match to it must be a false positive. The decoy hit count therefore estimates the number of false positives hiding among the target hits.

Why the factor of 2: the target and decoy databases are the same size, so the false positives among target matches should be roughly equal to the decoy matches — but reported FDR is normally quoted over all reported hits (targets plus decoys), giving the standard correction of multiplying by 2.

Why this matters more than it looks: a proteomics experiment performs tens of thousands of spectrum-to-peptide tests at once. Without this correction, a naive p-value threshold would admit a large number of wrong identifications — the same multiple-testing problem that requires BH-adjusted p-values in transcriptomics (L5).

Why it is characteristic of biologics: high-affinity binding plus a target present at appreciable levels means the elimination route is not only renal or proteolytic catabolism — it is consumption by the target itself. This is most visible with antibodies against membrane receptors or against abundant soluble ligands.

The PK signature: clearance is non-linear and dose-dependent. At low dose the target sink dominates, so increasing dose produces a more than proportional rise in exposure; at high dose the target is saturated, the sink is full, and PK becomes linear again. If exposure-versus-dose curves are non-proportional, suspect TMDD before suspecting assay error.

Dosing consequence: it explains why some antibodies need a loading dose to fill the target sink before maintenance dosing behaves predictably — the sink must be saturated before exposure scales normally.