Motif Detection in Yeast

VishakhJoe Bertolami

Nick UrreaJeff Weiss

Overview1. Problem Statement2. Motivation3. History4. Our Approach5. Evaluation6. Results7. Discussion8. References

1. The Problem Find regulatory sequences in the upstream

region of yeast DNA. Regulatory sequences are segments of

DNA where proteins can bind to enhance transcription of a gene.

The Problem We are given:

Upstream Genome- consists of: Gene Families- consists of:

Individual Genes- consists of: Strings like ATGC

We had to find substrings unusually frequent in gene families given their distribution in the whole upstream genome.

The Problem We emulated techniques devised by van

Helden. Worked on similar data set and tried to

emulate and even better his findings.

2. Motivation Organisms like yeast share many genes

with humans. As a result, they share diseases too. Finding regulatory sequences in yeast

might lead to medical advances. Might lead to therapies for diseases such

as cystic fibrosis.

3. History Previous century saw rapid advances in

genetics. Scientific community trying to get a better

understanding of various genomes. This particular technique was developed

by Jacques van Helden.

4 .Our approach Extract all substrings of lengths 6-8 in the

upstream genome. Calculate frequency of occurrence of each

substring. Put this data in a table.

Our Approach Consider a gene family. Find all substrings in it and frequencies

and build table. For each entry, add the probability of

occurrence. Use above data to calculate three scores.

Our Approach Score 1: Expected Occurrence / Actual

Occurrence Use probability of occurrence and size of

gene family to calculate expected occurrence.

Divide by actual occurrence. Low score -> Unusually frequent substring.

Our Approach Score 2: Poisson Distribution Use expected and actual number of

occurrences. If substring occurs ‘n’ times, calculate

probability of ‘n’ occurrences using Poisson Distribution.

Lower probability -> Unusually frequent

Our Approach Score 3: Binomial Theorem Use probability of occurrence, sizes of

genome and gene family and actual occurrences.

If substring occurs ‘n’ times, calculate probability of ‘n’ occurrences using Binomial Distribution.

Lower probability -> Unusually frequent

Our Approach Sort substrings by a score. Take top sequences, create a probability

matrix. Iterate probability matrix to get

probabilistic model of regulatory sequence.

5. Evaluation Metrics Van Helden’s results in ’98 paper and his

website. ’98 paper used old data, not very reliable

for evaluation. Website very useful since it works on

current data and dynamically calculates results.

Compared our output to his.

Evaluation Metrics Also, compare three scores types to find

best method.

6. ResultsComparison of Results for MET FAMILY

Gene Van Helden’s site Binomial Dist Poisson Dist Expected / Actual Old Paper

CACGTG 1 1 3 4 1

ACGTGA 2 2 1 2 3

TCACGT 3 3 2 1 2

ATATAT 4 4 N/A N/A 5

TATATA 5 5 N/A N/A 10

AACTGT 6 7 4 28 4

ACAGTT 7 6 N/A 29 N/A

ACACAC 8 9 7 N/A N/A

GTGTGT 9 8 6 N/A N/A

Results

Probability matrices generated successfully!

7. Discussion Paper results clearly outdated. Close co-relation with van Helden’s site. Binomial distribution best, followed by

Poisson and Expected/Actual

Discussion Why don’t Binomial results perfectly

match van Helden’s site? Van Helden paper only outlines general

method. He uses many filters and adjustments. Limited info about them on site. We used similar, but not same, filters. Example: Purge sequences that appear twice in

a row.

Discussion Future work

Find more filters. Try other similar organisms’ genomes. Biologically verify results!

Discussion What we learnt

Biology! First-hand look at genetic data Became more familiar with genes Clearly understood what the fuss about genetics is

about Computer Science

Teamwork Interfacing CS with other scientific disciplines

References van Helden, J., André, B. & Collado-Vides, J.

(1998). Extracting regulatory sites from the upstream region of yeast genes by computational analysis of oligonucleotide frequencies. J Mol Biol 281(5), 827-42.

van Helden, J., Rios, A. F. & Collado-Vides, J. (2000). Discovering regulatory elements in non-coding sequences by analysis of spaced dyads. Nucleic Acids Res. 28(8):1808-18.