03 - DNA Signals

DNA motifs = short, recurring patterns in DNA

• presumed to have biological function

• often indicating sequence-specific binding sites for proteins —> nucleases and transcription factors

• if you have all TFs/motifs

• —> global genetic regulatory network

• ribosome binding

• mRNA processing (splicing, editing, polyadenylation)

• transcription termination

• What turns genes on/off

• When a is gene regulated

• Where (in which cells) are genes regulated

• How many copies of the gene product are produced

Prokaryotic motifs:

• longer

• fewer transcription factors

• immediate upstream regulation

Eukaryotic motifs:

• shorter,

• more transcription factors per gene

• involve long-range regulation with much more noncoding sequences

• often form dimers or tetramers

• bind to palindromic sites

• combinatorial regulation (esp. eukaryotes)

  • order important

  • site spacing important

• from 1 genome:

  • Sequence overrepresentation

• functional genomics

  • predict regulons

• from N genomes:

  • phylogenetics footprinting

• from N genomes + funct. genomics

  • e.g. Phylocon

• degenerate consensus sequences

• sequence logos

• energy-normalized logos -> adjust for different GC-contents

• frequency of each nucleotide at every position in a motif

• scaled relative to the information content, showing conservation at each base.

The assumption that all four bases occur equally often is not always true, especially in organisms with biased GC content.

—> adjusts for low GC content by making certain bases, like G in a low-GC genome, carry more information, reflecting their significance

The tendency to overestimate the information content in small samples by not assuming that all bases are equally likely.

• additional data assumed to occur at least once in each alignment position

• —> overcome the lack of data and provide a more accurate probability estimate

• can find all possible sequences matching a pattern

• but: do not distinguish between consensus sequences and unlikely sequences

• model sequence data

• predicting the most likely state path (e.g., exon, intron)

• helping identify motifs like splice sites

• HMM states are multiplied —> results in very low numbers

• log probabilities and sum —> much easier numbers to work with

—> higher score = more likely to not be background

The Gibbs sampler is a stochastic method that iteratively samples subsequences to identify the most fitting motif model probabilistically.

• Expectation:

  • estimate the probability of finding the site at any position of the sequences

• Maximization: update expected base distributions

• Repeat until convergence

e.g., MEME does this

Phylogenetic footprinting identifies regulatory elements by comparing orthologous sequences from different species to find conserved regions.

The problem involves finding a set of l-mers in DNA sequences that maximizes the consensus score, given a sample of DNA sequences and the length of the motif.

A scoring function is used to evaluate and compare different guesses of motif starting positions to determine the best profile and consensus sequence.

The brute force approach computes scores for all possible combinations of starting positions to find the best motif, but it is computationally impractical due to the large number of possibilities.

Deterministic optimization involves methods like EM that systematically refine motif models based on weighted averages of probabilities until convergence.

The Gibbs sampler initializes with random motif sites, iteratively updates the motif model, and probabilistically selects new sites based on calculated weights until the best pattern is found.

Mutations can cause motifs to vary slightly between genes, complicating the identification of the consensus sequence and requiring algorithms to account for these variations.

Consensus sequences serve as a reference from which mutated motifs emerge, helping identify motifs by minimizing the distance between real motifs and the consensus.