BIOSC 1540: L05 (Gene annotation)

Computational Biology

(BIOSC 1540)

Sep 10, 2024

Lecture 05:
Gene annotation

Announcements

A01 solutions are posted
- Grading will take me a hot minute
A02 is due Thursday at 11:59 pm
Programming+ problems will be posted each homework and are completely optional

After today, you should be able to

1. Explain the graph traversal and contig extraction process in genome assemblers.
2. Understand key output files and quality metrics of genome assembly results.
3. Define gene annotation and describe its key components.
4. Outline the main computational methods used in gene prediction and annotation.
5. Analyze and interpret basic gene annotation data and outputs.

Walking along the graph produces strings

Results in CGTAAAT

CG

GT

TA

AA

AT

AAT

ATG

GGC

GCG

TGG

CGT

GTA

TAA

AAA

AAG

CGA

GAA

AGG

2

1

2

1

Results in: AATGGCGTAAAGGCGAA

Graphs in practice are not this easy

Graph traversal algorithms are used to extract contigs

General overview

Select a start node
Walk along the graph until a dead end or previously visited node is reached
Backtrack and explore alternative paths
Repeat for remaining unvisited nodes

Multiple approaches are used and comes down to personal preference

How do we select a starting node?

High coverage: Suggests that the node is likely a true sequence rather than an error

Hubs: Indegree and outdegree != 1

Hubs are shown as filled-in nodes

Hub

Not a hub

Depth-first search explores graph for potential paths (i.e., contigs)

Start at a chosen vertex (node).
Mark the current vertex as visited
Explore an adjacent unvisited vertex
If no unvisited adjacent vertices exist, backtrack to the last vertex with unvisited adjacent vertices.
Repeat steps 2-4 until all reachable vertices have been visited.

How do you choose a walk?

How do we choose the "best" path for our contig?

What factors would you look for?

Talk to your neighbors

Long paths are desired but not always reliable due to potential repeats

High, consistent read coverage

Unique, non-branching paths

After today, you should be able to

1. Explain the graph traversal and contig extraction process in genome assemblers.
2. Understand key output files and quality metrics of genome assembly results.
3. Define gene annotation and describe its key components.
4. Outline the main computational methods used in gene prediction and annotation.
5. Analyze and interpret basic gene annotation data and outputs.

Let's get practical with SPAdes

SPAdes is a popular prokaryote genome assembler

GitHub

Based on De Bruijn graphs with numerous improvements

Error correction with BayesHamming

Build Hamming graphs for k-mers

Undirected edges for Hamming distance of n nucleotide differences

Identify strong k-mers based on clustering (i.e., high similarity)

Estimate read error based on base qualitites

Builds multisized graphs with different k's

Leads to fragmented graphs and helps reduce repeat collapsing

Collapsed, tangled graphs great for low-coverage regions

By using multiple graphs, SPAdes can better handle variable coverage

Large k

Small k

Graph simplification and correction

Potential bulge

Removal of a bulge will quickly deteriorate the graph and lose read information

If P needs to be removed, we "project" the information (e.g., coverage) onto Q

P's edges are then removed in the process

Potential tips

Removes P (shortest) and projects information onto Q

Clarification: Paired-ended reads do not always cover our whole insert

Read 1 (forward) and Read 2 (reverse) are stored in FASTQ

If our insert (i.e., DNA sample) is longer than reads, then we don't sequence the inner distance

Should we minimize this inner distance?

False

A gap between paired reads gives us insight into repeated regions

Read 2

ATATATATATATATATAT

Read 1

ATATATATATATATATATAT

Gap

ATATATATATATATATATATATATATATATATATAT

Suppose I have an "AT" repeat for both Read 1 and 2

The assembler will have to figure out if these are overlapped or separated, but by how far?

Having a gap tells me they don't overlap, but for how long?

Knowing length of Read 1, Read 2, and total insert length allows me to calculate gap length

Assembly algorithms (e.g., SPAdes) can estimate this and refine their results

Assemblers provide contigs and scaffolds

We can visualize this using an assembly graph from a tool called Bandage

Contigs

Scaffolds

Each island contains one or more contigs

Each solid line is called a "node" (Why? I have no idea.) and represent a contig

suggests how these contigs connect to form a scaffold

connection

Each

After today, you should be able to

1. Explain the graph traversal and contig extraction process in genome assemblers.
2. Understand key output files and quality metrics of genome assembly results.
3. Define gene annotation and describe its key components.
4. Outline the main computational methods used in gene prediction and annotation.
5. Analyze and interpret basic gene annotation data and outputs.

Annotation is identifying the genetic elements and function in our contigs

Structural annotation identifies critical genetic elements such as genes, promoters, and regulatory elements

Functional annotation predicts the function of genetic elements

Eukaryote annotation is significantly more challenging that prokaryote

Introns and alternative splicing complicate annotation

Ab initio annotation for prokaryotes is tractable

We will focus on prokaryotes because eukaryotes are way more complicated

Example: Prokka

Example: AUGUSTUS

Prokaryotes

Eukaryotes

Probabilistic models to identify open reading frames

Accuracy demands supporting evidence like mRNA sequencing

After today, you should be able to

1. Explain the graph traversal and contig extraction process in genome assemblers.
2. Understand key output files and quality metrics of genome assembly results.
3. Define gene annotation and describe its key components.
4. Outline the main computational methods used in gene prediction and annotation.
5. Analyze and interpret basic gene annotation data and outputs.

Identify open reading frames (ORF)

Prokka uses prodigal for this

(I will use different notation than the paper.)

Seek the standard start codons: ATG, GTG or TTG

Seek stop codons based on the translation table

TAA, TAG, and TGA for bacteria, archaea, and plant plastids (Code 11)

Score potential ORFs

S_\text{ORF}(n) = 4.25 \left[ S_R(n) + S_T(n) + 0.4 S_U (n) \right] + S_C(n)

Ribosomal binding site motif score

Start type score

Upstream score

Coding score

S_R(n)

S_T(n)

S_U(n)

S_C(n)

RBS score computed from dataset fitting

S_R(n) = \log \left( \frac{R(n)}{B(n)} \right)

Took training data from 12 annotated genomes

Computed frequency of RBS motif bin in

R(n)

Entire sequence
(Baseline)

RBS frequency

B(n)

Search for RBS motif after start codon; choose whichever has the lowest bin number

Start

Spacer

RBS

Start codon score given by similar RBS framework

S_T(n) = \log \left( \frac{T(n)}{B(n)} \right)

Took training data from 12 annotated genomes

Computed frequency of start codon in

T(n)

Entire sequence
(Baseline)

Start codon frequency

B(n)

Upstream score based on base analysis

Start

Stop

-2 to -1

-44 to -15

By analyzing base frequency in specific upstream regions, their annotation results improved

S_U(n) = w_\text{start} \sum_{i \in P} p_i \left(\text{nuc}_i \right)

Essentially looking for promotors

Coding score computed based on gene enrichment parameters

ATGGCC

CAGCTG

GGGCCC

ACTAGT

Example hexamers called "words"

Computed frequency of nucleotide hexamers called "words" in

G(w)

Compute probability of observing word within the whole genome

within genes

B(w)

Compute the probability of observing word

C(w) = \log \left( \frac{G(w)}{B(w)} \right)

Word coding score

Coding score computed based on gene enrichment parameters

It can be thought of as
"How often does this word appear in genes?"

Gene coding score

S_C (n_\text{start} \ldots n_\text{stop}) = \sum_{i=n_\text{start}}^{n_\text{stop}} C\left( w (i) \right)

Sum hexamer word score and shift over one codon at a time

1

2

3

ATGCATGCTTAG

Results: Sequences that likely encode for proteins

Potential protein

Non-coding

Functional annotation is normally based on protein database search

Similarity search will be our topic for Thursday

After today, you should be able to

1. Explain the graph traversal and contig extraction process in genome assemblers.
2. Understand key output files and quality metrics of genome assembly results.
3. Define gene annotation and describe its key components.
4. Outline the main computational methods used in gene prediction and annotation.
5. Analyze and interpret basic gene annotation data and outputs.

Prokka will provide several outputs

>ECNNONJI_02637 Dihydrofolate reductase
MTLSILVAHDLQRVIGFENQLPWHLPNDLKHVKKLSTGHTLVMGRKTFESIGKPLPNRRN
VVLTSDTSFNVEGVDVIHSIEDIYQLPGHVFIFGGQTLFEEMIDKVDDMYITVIEGKFRG
DTFFPPYTFEDWEVASSVEGKLDEKNTIPHTFLHLIRKK