l529 - presentation proteomics - yogita mantri -arvind gopu 11/10/2003
TRANSCRIPT
L529 - Presentation
PROTEOMICS
- Yogita Mantri -Arvind Gopu
11/10/2003
Introduction – What is Proteomics?
“The identification, characterization and quantification of all proteins involved in a
particular pathway, organelle, cell, tissue, organ or organism that can be studied in concert to
provide accurate and comprehensive data about that system.”
http://www.inproteomics.com/prodef.html
Central lesson from eukaryotic genome projects
Evolutionary complexity is not primarily determined by increasing the number of genes, but by increasing variation on the level of the synthesized proteins.
This is achieved by generating MULTIPLE proteins from only ONE gene e.g. by different combinations of exons by alternative splicing post-translational protein processing (e.g. cleavage of pro-
peptides) post-translational protein modifications (e.g. acetylation,
glycosylation) modified central dogma: DNA --> RNA --> protein(s) it is important to perform analyses on the level of gene
PRODUCTS Key
Key advantage of proteomics Researchers work on the level of gene products and
deal with genes that are really expressed to give a detectable PRODUCT and are not just "expressed“ which only says they produce a detectable mRNA but it is not clear whether there is a gene product or not.
Key limitation of proteomics Usually, only a fraction of the proteins synthesized can
be detected in a proteomics experiment, whereas the expression of ALL genes can be monitored in a whole-genome array experiment.
Key prerequisite of proteomics A genome sequence for the investigated organism or at
least a collection of many cDNA sequences is required.
Experimental Background
Mass Spectrometry
What is Mass Spec?
Analytical tool measuring molecular weight (MW) of sample
Only picomolar concentrations required Within an accuracy of 0.01% of total weight of
sample and within 5 ppm for small organic molecules
For a Mr of 40 kDa, there is a 4 Da error This means it can detect amino acid substitutions /
post-translational modifications
What sort of info is returned?
Structural information can be generated Particularly using tandem mass spectrometers Fragment sample & analyse products Useful for peptide & oligonucleotide sequencing Plus identification of individual compounds in
complex mixtures
How does a Mass Spectrometer work?
3 fundamental parts: the ionisation source, the analyser, the detector
Samples easier to manipulate if ionised Separation in analyser according to mass-to-charge
ratios (m/z) Detection of separated ions and their relative
abundance Signals sent to data system and formatted in a m/z
spectrum
Simplified Schematic
The analyser, detector and ionisation source are under high vacuum to allow unhindered movement of ions
Operation is under complete data system control
Schematic of a typical TOF-MS/MS
Sample Introduction& Ionisation
Direct into ionisation source or via chromatography for component separation (HPLC, GC, capillary electrophoresis)
Ionisation can be positively charged (for proteins) or negatively charged (for saccharides and oligonucleotides)
Ionisation methods Atmospheric Pressure Chemical Ionisation (APCI) Chemical Ionisation (CI) Electron Impact (EI) Electrospray Ionisation (ESI) Fast Atom Bombardment (FAB) Field Desorption / Field Ionisation (FD/FI) Matrix Assisted Laser Desorption Ionisation
(MALDI) (Clemmer Group) Thermospray Ionisation
Detection & Recording of Ions
Detector monitors ion current, amplifies it and then transmits signal to data system
Common detectors: photomultiplier, electron multiplier, micro-channel plate
Mass spectrometry is a very powerful method to analyse the structure of organic compounds, but suffers from 3 major limitations:
Compounds cannot be characterised without clean samples
This technique has not the ability to provide sensitive and selective analysis of complex mixture
For big molecules like peptides spectra are very complex and very difficult to interpret
Tandem MS or MS/MS has 2 mass spectrometers in series.
In first mass spectrometer (MS1) is used to SELECT, from theprimary ions, those of a particular m/z value which then pass intothe Fragmentation Region. The ion selected by the MS1 is the parent ion and can be a molecular ion resulting from the primary fragmentation. DISSOCIATION occurs in the fragmentation region. The daughter ions are analysed in the Second Spectrometer (MS2). In fact, the MS1 can be viewed as an ion source for MS2.
MS2MS1
Peptide Sequencing Peptides of 2.5 kDa or less give best data Protein sample often taken from 2-D gels and digested A protein digest can be analysed as entire mix Initial MS spectrum showing Mr of all components in digest
(peptide map) may be enough for a database search and identification
Peptides fragmented along the amino acid backbone in tandem mass spectrometry
Some peptides generate enough info for full sequence, others only generate partial sequences of 4-5 amino acids
Often this “tag” sequence is sufficient for database identification
Data Analysis
Common Data Analysis - Pipeline
Issue #1 (Relatively Minor?)
Diverse set of Mass Spectrometers… More flexibility BUT ... Different data formats Limited Data analysis possible Exchange of RAW datasets and creation of public
repositories for the data/software? Not easy if not impossible
Work Around for Issue #1?
To get around this problem Convert to ASCII text - speed and loss of precision can be
an issue Other formats specific to this field A lot of XML based file formats seem be floating around Of course using XML format (for example) gives raise to
additional level of complexity -- parsers, formatters, etc It does add flexibility between data formats Indexing techniques used to speed up access
Issue #2 (Much bigger!)
Data Size Higher Dimensionality The combination is even deadlier!
More detail in a minute … Before that … The LC/MSMS spectrum data looks like this:
LC Drift TOF Intensity i.e., 3-D + Intensity
Issue #2 (Continued…)
As a first step in data analysis: Find peaks in the LC/MSMS data
Peaks is kind of a misnomer. Center of mass (or something like that) is a better term. Illustrates inherent non-uniformity within proteomics circles Easier said than done as we found out!
Let us start with a simpler case of finding peaks in 2-D data – a little more complicated than 1-D …
Peak Finding – 2-D data
http://www.cs.nott.ac.uk/~gxk/aim/notes/hillclimbing.doc
Peak Finding - Higher Dimensions?
As mentioned earlier data is of the form: LC Drift TOF Intensity i.e., 3-D + Intensity
Add to this huge data size and get a hang of how difficult a problem it is
Some Possible Solutions
Solutions we thought about: Find peaks using a brute force approach
Not computationally feasible in terms of time and memory
Squeeze 3-D data into 2-D, find peaks and then work backwards. This is the algorithm implemented by Frank - one of the
IU Chemistry folks Use existing implementations of graph functions
available in packages (For example: LEDA) to preprocess data and then find peaks on smaller data set
Our Peak Finding Algorithm
Used LEDA package for C++ Specifically made use of O(n Log n) implementation
of Delaunay Triangulation Neighbor Finding algorithm in 3 D space
Once neighbors were found then do a brute force peak finding step How good were our results?
More details? Take a look at our summer presentation at Chemistry
Sample of the data … What it looks like?
Peptide Assignment
Find sequence of amino acids that can generate the list of masses seen in the tandem MS scan.
Many different strategies: Searching MS/MS spectra against a sequence
database (Sequest, Mascot, etc) De novo sequencing (no database!) Hybrid
Scoring Peptide Sequences
Multiple search engines are available Sequest and Mascot
They use different scoring algorithms Search outputs are not comparable Search outputs usually require expert
validation …
An example of scoring system: SCOPE
Probabilistic model for scoring tandem MS against peptide database
Two stage model Uses dynamic programming Incorporates fragment ion probabilities, noisy
spectra and instrument measurement error Details:
http://bioinformatics.oupjournals.org/cgi/screenpdf/17/suppl_1/S13.pdf (Scoring Spectra section)
Peptide Validation
Validate peptide assignments made during the database search step.
Obviously, method used should be standardized and independent from the experimental and computational methods used
Manual Validation
Filtering by database search scores Problems:
Filtering criteria vary among researchers Error rates are unknown Possible only on very small datasets
Model Based Validation?
Empirical Statistical Model to estimate accuracy … Anal. Chem 2002, 74, 5383 – 5392
Employs Expectation Maximization and Machine Learning techniques
Learns to distinguish between correct from incorrect database search results
Model Based Validation – EM algorithm
Each peptide assignment evaluated w.r.t. all other assignments including incorrect ones
Denote correct and incorrect assignments as (+) and (-); Scores as x_1, x_2 … x_s P(+ | x_1, x_2 … x_s) =
P(x_1, x_2 … x_s | +) * P(+)
---------------------------------------
∑ P(x_1, x_2 … x_s | i) * P(i)
Model Based Validation – EM algorithm (Continued …) Replace search scores with discriminant
function F P(F| +) * P(+)
P(+ | F) = -------------------
∑ P(F| i) * P(i) Bunch of probabilistic parameters considered Ended up approximating distributions to
Gaussian and Gamma distrs. (More details out of scope of this presentation, please refer paper)
Example of Automated Validation
An example: Protein Prophet Compute probabilities that peptides assigned to
MS/MS spectra are correct Learns distributions of search scores and peptide
properties among correct and incorrect results The computed probabilities are claimed to be a true
measure of the confidence! Combines probabilities of peptides assigned to
MS/MS spectra to compute probability that corresponding proteins are present in the sample
Interpretation
Assign a biological meaning to the output of the pipeline
Current Issues and Challenges
Slide adapted from http://www.ciphergen.com/tech_doc11.2.html
After Proteomics…..
Functional Genomics
ProteinChipTM.
Limitations of Proteomics
Experimental limitations:Large-scale protein analysis difficult because:
-Proteins are fragile
-They can exist in multiple isoforms
-There is no protein equivalent of PCR for amplification of a small sample
Data Analysis Limitations:-Data contains a lot of noise that is difficult to separate from actual signal. This results in wastage of computing resources on searching for unlikely spectra.-Database searches for matching spectra only give scores, leaving manual intervention necessary for eliminating false positives
Biomedical limitations
-In practice, it is very difficult to trace the complete progression of a disease.
-Hence, using proteomics for monitoring the biochemistry of a disease is like using a photo camera to record a football match.
-Case of breast cancer research:
http://www.mcponline.org/cgi/reprint/2/5/281.pdf
References and Further Reading
Explains the whole process nicely -- articlehttp://swehsc.pharmacy.arizona.edu/analysis/Proteomics_News.htm
Mascot Home page -- help sectionhttp://www.matrixscience.com/help_index.html
Presentation about MS MS datahttp://sashimi.sourceforge.net/extra/oral.pdf
http://www.genetik.uni-bielefeld.de/D1E33C76A7CCA010AAD3B435B51404EE/Genome_Research_WS2002_03/stunde_ws0203_10.pdhttp://bmbus6.leeds.ac.uk/mres/5130/MassSpectrometry.ppt
Some info about drug discovery/economic issues n such:http://monod.uwaterloo.ca/cs798/proteomics.pdf
Paper on interpreting MSMS data http://chem-ncms.unl.edu/asms2003/kurt.pdf
How to estimate correctness of MS MS prediction -- EM !!!http://www.proteomecenter.org/PDFs/Keller.AnalChem.02.pdf
http://www.nature.com/cgi-taf/DynaPage.taf?file=/nbt/journal/v21/n3/full/nbt0303-221.html
http://www.esainc.com/MolecularProteomics/molecular_proteomics.htm
Others:http://genome.ucsd.edu/classes/be202/ppt/11
Delaunay Triangulation:http://almond.srv.cs.cmu.edu/afs/cs/project/quake/public/www/triangle.delaunay.html
SCOPE paper -- screen PDFhttp://bioinformatics.oupjournals.org/cgi/screenpdf/17/suppl_1/S13.pdf
Internet sites
www.astbury.leeds.ac.uk/Facil/MStut/mstutorial.htm(Dr Alison E. Ashcroft at Leeds)
www.asms.org (The American Society for Mass Spectroscopy) www.spectroscopynow.com (Base Peak)
Mass Spec tools www.expasy.ch/tools/#proteome http://prowl.rockefeller.edu www.mann.embl-heidelberg.de
Bibliography
Internet sites :
http://www.google.com• http://www.bmss.org.uk/what_is/whatis.html• http://www.duke.edu/~mdfeezor/NSHome/inform/msms1.html• http://www.astbury.leeds.ac.uk/Facil/MStut/mstutorial.htm• http://ms.mc.vanderbilt.edu/tutorials/ms/3.htm • http://www.garvan.unsw.edu.au/public/corthals/book/IPMS.html• http://www.micromass.co.uk/basics/Glossary.html
Ionization Methods
Further Reading1. For MALDI beginner:http://www.srsmaldi.com/Maldi/Guide.html
2. For MALDI lab user:
http://www.srsmaldi.com/Maldi/Lab.html 3. For MALDI tutorial:
http://ms.mc.vanderbilt.edu/tutorials/maldi/maldi-ie_files/frame.htm 4. Ionization Methods 1:http://www.jeol.com/ms/docs/ionize.html
5. Ionization Methods 2:http://www.waters.com/Waters_Website/Applications/lcms/lcms_itq.htm
SELDI Web sites:
• Molecular Analytical Systems (MAS).http://www.seldi.org/
• Manufacturers of ProteinChip(R)http://www.ciphergen.com/