Using phylogenetic profiles to predict protein function and localization

1
Using phylogenetic profiles to predict protein
function and localization

• As discussed by Catherine Grasso

2
Papers

• Pellegrini, et al. Assigning protein functions
  by comparative genome analysis Protein
  phylogenetic profiles. (1999) PNAS 96, 4285-4288.
• Marcotte, et al. Localizing proteins in the cell
  from their phylogenetic profiles. (2000) PNAS 97,
  12115-12120.

3
Basic Idea

• Sequence alignment is a good way to infer protein
  function, when two proteins do the exact same
  thing in two different organisms.
• Proteins with gt 30 sequence identity have the
  same fold, and typically the same function.

4
Basic Idea

• But can we decide if two proteins function in the
  same pathway, such as histidine biosynthesis, or
  the same biomolecular structure, such as the
  flagella or ribosome, even if they dont do the
  exact same thing?
• Yes. Assume that if the two proteins function
  together they must evolve in a correlated
  fashion so every organism that has a homolog of
  one of the proteins must also have a homolog of
  the other protein.

5
Phylogenetic Profile

• For a given protein, BLAST against N sequenced
  genomes.
• Construct a vector with N coordinates.
• If protein has a homolog in the organism n, set
  coordinate n to 1. Otherwise set it to 0.

Protein P1 0 0 1 0 1 1 0
0
6
Functional Link

• Assign a degree of functional linkage between P1
  and P2 based on the number of positions (or bits)
  at which their profiles differ.

Protein P1 0 0 1 0 1 1 0
0
Protein P2 0 1 1 0 1 1 0
0
7
What They Did

• Computed phylogenetic profiles for 4,290
  proteins in E. Coli.
• Aligned each protein sequence Pi with the
  proteins from 16 other fully sequenced genomes.
• Proteins coded by genome n are defined as
  including a homolog of Pi if they align to Pi
  with a score that is deemed statistically
  significant.

8
Conclusions

• Comparing profiles is useful tool for identifying
  the complex or pathway in which a protein
  participates.
• As the number of fully sequenced genomes
  increases scientists will be able to construct
  longer more informative profiles.
• In 1999, 100 more genomes were due to be
  completed in next few months.
• Suggests that as eukaryotic genomes come out
  profiles will be a useful tool for studying
  pathways in higher organisms.

9
Evolutionary Origin of Eukaryotic Cell

• Mitochondria, chloroplasts and perhaps other
  organelles descended from microbes captured by
  progenitors of eukaryotic cells.
• You exist because of a bad case of indigestion!

10
Evolutionary Origin of Eukaryotic Cell

• This endosymbiosis was stabilized by shifting of
  genes of organelle into nuclear genome and
  transport systems being established to shuttle
  organellar proteins form cytoplasm into
  organelles.
• Contemporary mitochondrial genome encode only a
  few genes (lt20), primarily large integral
  membrane proteins which cant be transported.

11
Evidence

• Proteins of these organelles have molecular
  properties resembling prokaryotic rather than
  eukaryotic proteins

1. Average lengths
2. Domain composition
3. Amino acid composition
4. Homologs among prokaryotes

12
Phylogenetic profiles

• Will show that proteins with similar phylogenetic
  profiles localize to similar subcellular
  locations.
• Actually, will primarily show this for the
  mitochondria.

13
Calculating phylogenetic profiles

• In this study, the value at each position of the
  profile is equal to -1/log E, where E is the
  BLAST expectation value of best matching protein
  in a genome.
• Calculated only for E lt 1x10-6 and 1.0 otherwise.
  So zero is a perfect match and one is no match.

14
Three Categories

• Prokaryote Derived Only has homologs in
  prokaryotes.
• Eukaryote Derived Only has homologs in
  eukaryotes.
• Organism Specific Has no homologs.
• Why split these categories? Should have
  different functions and roles in mitochondria.

15
Linear Discriminant Functions
t
Varying t increases prediction accuracy at the
expense of coverage.
MP
Non-MP
16
Testing Algorithm

• First, predicted the location of yeast proteins
  of known location (open diamonds).
• Second, a jackknife test was performed. Repeated
  100 times with different random sets (filled
  diamonds). Coverage 58 at 50 accuracy.
• Third, used yeast proteins as training set and
  worm proteins as test set. Coverage 65 at 50
  accuracy.

17
Prediction

• Applied algorithm to all yeast proteins.
  Estimate 630 total mitochondrion-targeted genes
  in yeast or 10 of genome.
• Applied algorithm to all worm proteins. Estimate
  660 total mitochondrion-targeted genes in worms
  of 4 of genome.

18
Verifications

• Tested whether functions of newly predicted
  mitochondrial proteins matched functions of known
  mitochondrial protein better than the functions
  of a random set of proteins. (Jacard
  Coefficient, Pie Charts)
• Fraction of predicted mitochondrial proteins with
  predicted transmembrane segments or signal
  peptides.
• 2D gel of whole rat liver and human placental
  mitochondria reveals 250-350 visible proteins.

19
Conclusions

• There is information in the phylogenetic
  profiles, but it is quite noisy.
• Yields approximate numbers of genes migrated to
  the nuclear genomes from the mitochondria.
• Gives even more evidence for endosymbiotic
  theory.
• However, verifications did not confirm results as
  much as one might like.
• Perhaps fundamental assumption flawed.