Important questions

• Documentation: comparative genomics for documenting different species.

• Use: identifying natural products with medical relevance.

• Protection: identifying resistance genes in beech treees

Why do we sequence?

To read genetic information so we can find genes, detect disease variants, understand how organisms work and adapt, compare genomes, and uncover evolutionary relationships for medicine, agriculture, and biotechnology.

What does sequencing mean?

Determining the exact order of building blocks in a molecule—most commonly the order of DNA bases (A, T, G, C).

Influence of sequencing initiatives on biodiversity?

They’ve accelerated species discovery, enabled rapid monitoring via DNA barcoding and environmental DNA, provided reference genomes revealing adaptations and population health, and directly strengthened conservation decisions by identifying genetic vulnerabilities.

• An organism/tissue can be examined under different experimental conditions, e.g. which gene is upregulated or downregulated.

• A complete genome, including regulatory regions and introns, is not important if you are only interested in genes/proteins.

• Reference-based approach.

• De novo approach.

• The k-mer size should not be too small so that parts of the sequence that occur more frequently can be distinguished from each other.

• For repetitive patterns, should be larger than the repetitive pattern.

• Ideally choose so that each k-mer occurs only once.

• k-1 gives the maximum possible overlap of the k-mers, which increases specificity.

• If k = 1, the contig could not be extended and would consist only of k-mers.

• If k > 1, errors can occur.

• FASTQ: sequence plus quality values.

• FASTA: sequence only, as a file format.

• Collect and select individual pieces of information.

• Combine and integrate information into a database.

• Present the information, e.g. through a GUI/API.

The database may be incomplete.

• Protein databases: linear growth, limited by laboratory work.

• Sequence databases: exponential growth, because technology becomes better and faster.

RefSeq and GenBank

• Intrinsic = information that can be extracted from the sequence itself.

• Extrinsic = information obtained by comparison with other sequences/models.

• Position-specific scoring matrices used to identify motifs in biological sequences (DNA, RNA, protein).

1. Training data.

2. Create a count matrix.

3. Add pseudocounts.

4. Calculate the Kullback-Leibler distance for each position.

5. The sum of the Kullback-Leibler distances for each position gives the PSSM score for the sequence.

6. Difference between a scoring matrix (BLOSUM62) and a PSSM.

• BLOSUM62 is position-independent: the score for substituting G with A is the same at every position.

• PSSM is position-specific: the score for substituting G with A depends on the position.

Cellular Component, Molecular Function, Biological Process.

• Assign GO terms to genes for functional annotation.

• Standardized terminology makes categorization possible.

• Each gene is described with specific GO terms; the path to get there stays the same, so less data has to be stored.

• HMMs can find motifs even when the length changes due to insertions/deletions, and motifs can be used to assign function to sequences.

• PSSMs are position-specific but only work with fixed length; insertions/deletions strongly change scores. PSSMs can still recognize motifs and assign function.

• BLOSUM62 is position-independent and gives no information about function.

• KEGG shows the organization of genes in different metabolic pathways.

• KEGG assigns each gene a KO term that functionally describes it.

• The database contains orthologs with experimentally characterized genes/proteins.

• KEGG Mapper creates metabolic pathways as a network of KO terms.

• Homology refers only to evolutionary relationships.

• Sequence similarity is not part of the concept of homology; it measures how much two sequences match.

• Sequences that are more similar than expected by chance are evolutionarily related, i.e. homologous.

• Experimental annotation.

• Electronic annotation, most common.

• Curated non-experimental annotation.

• Describing gene function.

• Restricted vocabulary.

• Hierarchical organization.

• Directed acyclic graph.

• 3 sub-ontologies.

• Evidence codes.

• Orthologs are homologous sequences that were separated by a speciation event.

1. Speciation event: reproductive isolation of two populations leads to separation of genetic lineages.

2. Isolated evolution of the sequences.

3. The resulting sequences are called orthologs.

• In reference-based assembly, a sufficiently well-annotated genome of the organism already exists. The reads can then simply be mapped to it. It is like a puzzle with the solution picture already available.

• De novo assembly joins overlapping sequence reads with sufficient similarity into longer sequences, also called contigs. The contigs reconstruct the original transcript. De novo assembly is a puzzle without a template.

• If a genome is available, reference-based assembly is always the better and easier approach. For transcript analysis, for example, different isoforms can be compared more easily with the reference-based approach. Split reads and exon-intron boundaries can also be identified, as well as which introns were spliced out. It can also reveal which transcripts are completely missing from the dataset.

• Which assembly approach do you use? Reference-based approach.

• What do you need for this? A reference genome.

• PacBio and Nanopore, because they provide long reads.

PacBio and Nanopore have the “problem” that only about 85% of reads are correct, so around 15% contain errors.

• PacBio reads often show insertions.

• Nanopore sequencing often shows deletions, especially when the same nucleotide repeats several times.

• Illumina sequencing is about 99% correct but can only generate short reads.

• Therefore Nanopore and Illumina are combined, and the sequences are compared. Nanopore provides long reads that are corrected by short Illumina reads.

• Many sequences can be sequenced in parallel.

• It is automatable.

• Read accuracy is high, about 99.3% per read.

• Many sequences can be sequenced in parallel.

• It is automatable.

• Read accuracy is high, about 99.3% per read.

• Only short reads can be generated.

• Sequencing quality decreases toward the end of the read.

• A cluster consists of many copies of a sequence. When sequencing starts, nucleotides are incorporated.

• It can happen that not all DNA polymerases in the cluster incorporate a nucleotide, even though they should have.

• As a result, these sequences are no longer synchronized with the cluster.

• Over the course of sequencing, these errors accumulate, so by the end of 125 sequenced bases the signal is no longer as good as at the beginning.

• An isoform arises in eukaryotes through alternative splicing, meaning it is a possible transcript of a gene.

• Multiple proteins can arise from one transcript.

• Cassette exons: different exons can be combined, but not every exon has to be present; this is exon skipping.

• Intron retention.

• Contig: a contig is a set of reads connected by overlap of their sequences. All reads belong to one and only one contig, and each contig contains at least one read. Contigs have no Ns.

• Scaffold: consists of ordered and oriented, but usually non-overlapping, contigs separated by gaps of approximately known length. Scaffolds are usually formed by identifying contig pairs that each contain a read from a read pair.

• We need information about how many Ns are between the individual contigs.

• Read-pair information is needed to connect contigs, since they do not have to overlap.

• A structured representation of something, highlighting relevant features and ignoring random similarities.

• It describes the variability between different instances of the modeled object.

• It helps classify new members of the set of candidates.

• It helps identify outliers that deviate from the average more than expected.

• Overfitting means that too many explanatory variables were used to describe a model.

• This makes the model too narrow, e.g. focused on one very specific pig breed, while others are ignored.

• A training dataset contains examples used to learn patterns and relationships in the data.

• The algorithm’s weights are adjusted using the training data, meaning the algorithm learns from them.

• By applying a PSSM to score them.

• If a long time has passed and conserved sections are still present, these are evolutionarily important sequence regions.

• They are functionally relevant, and functionally relevant regions have not changed over time.

• Scores the position-independent probability of amino acid substitutions.

• The model is derived from alignments of homologous sequences.

• This allows similarity between two sequences to be evaluated.

PSSM

• Position-specific.

• A model is generated from training data.

• A model for a specific motif.

• Amino acid occurrence at each position is described by its probability distribution, i.e. a count matrix.

• Sequence is compared with the model.

• This allows recognition of conserved patterns (motifs) in the sequence being compared.

• Length-independent.

• Position-specific.

• The model allows partial domains, duplications, and links.

• It is complex enough to cover all possible cases without overfitting.

• PSSMs are faster, but worse, because they are length-variable.

• GenBank is not curated, author-submitted, can contain multiple records for the same locus, and records can contradict each other.

• RefSeq is curated, created by NCBI from existing data, revised as new data emerge, and usually contains single records for major organisms.

Regions that have not changed over time because they are functionally relevant.

• QC.

• Trimming.

• Assembly.

• Assembly validation.

• Annotation, e.g. domains & motifs, orthology search, homology search.

• Base substitutions are most common in Illumina.

• Deletions are most common in Oxford Nanopore.

• Insertions are most common in PacBio.

• Errors are less serious when detecting conservation between species; at worst they weaken the signal, which may cause conserved regions to be missed. That is a false negative.

• If you compare different individuals of the same species, where explicit differences matter, reconstruction errors can make a huge difference. That is a false positive.

• Paralogous genes are similar in sequence but located at different positions in the genome.

• Paralogous genes arise through duplication events.

• Orthologous genes differ in sequence from each other but derive from the same ancestral gene.

• Orthologs arise through speciation events.

• By sequence similarity, comparing the entire sequence with all models in a database.

• If similarity is stronger than expected by chance, the sequence is assumed to be an ortholog.

• Homology means that two sequences share a common ancestor.

• It is not a statement about sequence similarity.

• It is a yes/no concept.

• Homology does not say anything about function.

• Sequences that are not related are not more similar than expected by chance.

• Sequences that are more similar than expected by chance are homologous.

• This does not mean that homologous sequences must always be more similar than chance.

• A position-specific but length-independent model for identifying conserved regions.

• Generated via a multiple sequence alignment, building on PSSMs.

• Position-specific columns are transformed into match states.

• Add delete states and insert states.

• Has start and end states and can represent repeated domains through joining states.

• A motif is usually a non-stable or not independently folding region.

• It presents extrinsic information in protein sequence analysis.

• Typically a short sequence region associated with a function.

• Fixed length, usually without gaps.

• Relatively high false-positive rate, because motifs are short and therefore likely to occur by chance.

• Represented by a regular expression or by a PSSM.

• PSSMs often include additional prior information for unobserved data.

• Commonly used to predict binding sites and modification sites.

• PSSMs can also be used to annotate domains quickly.

Jellyfish:

• Extract and count k-mers (k=25) from the reads.

• Store k-mers in a hash table with their frequencies.

• Sort by frequency.

Inchworm:

• Contig assembly from k-mers.

• Start at the top of the hash table, i.e. the most frequent k-mers.

• Search the hash table for k-mers that overlap with the chosen k-mer by .

• Each k-mer may only be used once.

• Extend the contig in 5’ and 3’ direction.

• Continue until no further extension is possible.

• Then take the next most frequent unused k-mer and assemble a new contig.

Chrysalis:

• Integrates contigs overlapping by , clustering them together, and extends using paired-end information.

• Assigns reads to the clusters that support the contigs.

• Converts the clusters into de Bruijn graphs.

Butterfly:

• Simplifies the de Bruijn graph.

• Records only where nucleotides vary.

• Checks which reads favor which path.

• These preferences can then be translated into different transcripts.

DNA polymerase cannot synthesize de novo; it needs something to attach to.