LargeScale Oligonucleotide Design: Principles and Application in Osprey Dr' Christoph Sensen

1
Large-Scale Oligonucleotide Design Principles
and Application in OspreyDr. Christoph Sensen
2
Overview

• Review of the Biochemistry
• Review of exisitng techniques and component tools
• How weve improved calculations in Osprey
• How to access Osprey
• Goals
• See existing analysis tools in a larger context
• Realize the advantages of computational
  expensive dynamic methods

3
Oligo Usage

• Osprey focuses on applications that require the
  design of large numbers of oligos, i.e. those
  that benefit from automation.
• Microarrays
• Large clone and directed genome sequencing

4
Thermodynamic Parameters

• We want
• Binding to target sequence that has the right
  duplex melting temperature for the experiment
• Distinguish between similar genes
• We want to avoid
• Oligos that fold back on themselves (hairpins)
• Oligos that bind to other copies of themselves
  (dimers)
• Oligos that bind well to more than one potential
  location in the DNA sample (secondary binding)

5
Melting Temperatures

• Oligo duplexes are not two state, therefore the
  melting temperature (Tm) is considered the point
  at which half the duplexes are denatured.
• Simple models include
• Wallace Tm 2(AT) 4(GC) (in 1M NaCl)
• Tm 100.5 41(GC)/(ATGC) - (820/(ATGC))
  16.6log10(Na)
• There are many models for determining melting
  temperature, but they share a common theme...

6
Oligonucleotide Length

• Melting temperature is directly proportional to
  oligo duplex length, and dependent on the base
  composition of the sequence.
• These two factors create a sweet spot for
  finding oligos of the appropriate melting
  temperature based on a oligo length base
  composition probabilities in the input sequence
• e.g. In Desulfolovibrio vulgaris (65 GC) the
  average 70mer melts at about 100C in 1M NaCl
• In Sulfolobus solfataricus (37 GC) it's about
  65C

7
Melting Temperature Complex

• More throrough models are based on physics
  melting temperature is determined by the enthalpy
  and the entropy of the oligo duplex
• Tm ?Hº/(?Sº R ln (C/4))
• Where
• ?Hº enthalpy (order) for the whole duplex
• ?Sº entropy (disorder) for the whole duplex
• R molar gas constant, 1.987 cal/(Kmol)
• C concentration of DNA

8
DNA Duplex Energy

• There are several models to determine enthalpy
  and entropy, all based on the accepted Nearest
  Neighbour (NN) concept the overall energy of the
  duplex can be predicted by summing the
  interaction of adjacent basepairs. e.g.
• Enthalpy (H) of GCCCTA
• H(GC)2H(CC)H(CT)H(TA)
• Neighbour energies are experimentally derived

9
NN Techniques

• Models derived from empirical data include
  Gotoh, Vologodskii, Breslauer, Benight, Sugimoto
• SantaLucia created a unified model from the
  combination of these datasets and his own data,
  generally considered the best model yet.

10
Popular Oligo Tools

• Primer3 (Uses Breslauer model, good for PCR
  primers, interactive usage)
• HyTher (SantaLucia's own system, accurate,
  interactive usage)
• No product has dominated the microarray design
  market yet, most chips are designed using
  commercial software
• Free software includes OligoArray, OligoChecker,
  PrimeGens

11
Why develop new tools?

• Many programs require intimate knowledge of
  parameters or cast a wide net to find oligos
• For large scale projects such as microarrays, the
  accuracy of the oligos can greatly affect their
  usability (e.g. ensuring even melting
  temperatures)
• Many tools can't deal with large secondary
  binding data (i.e., the known transcriptome) well.

12
Osprey Goals

• Require minimal user parameterization data
  preprocessing
• Deal nicely with large datasets
• Be as accurate as possible
• Be fast

13
General Procedure for Design

• Read in sequence
• Filter repetitive elements
• Determine optimal melting temperature/ oligo
  length
• Find candidate duplexes
• Check for undesirable oligo configurations
• Check for undesirable binding to the potential
  sample

14
Search Techniques

• Run the program iteratively, starting with strict
  constraints, and automaticallky relax them as
  needed, optimizing oligo similarity of length and
  melting temperature
• Why reinvent the wheel? There are techniques and
  tools that can be used and adapted to satisfy the
  oligo selection criteria
• These include heuristic methods such as BLAST
  for filtering, and dynamic methods where utmost
  accuracy is required

15
Repetitive Regions

• The design of oligos for degenerative repetitive
  regions can be tricky, but we would like to
  isolate the overlap regions (of significant
  length) to bind unique sets of genes.

16
Repetitive Analysis Tools

• MEGABLAST compares two large DNA sequences
  against each other. If the query and the subject
  are the same, you find repeats!
• We use this in Osprey, iteratively, to filter out
  repeats, leaving one copy of each repeat.
• The key parameters to tweak in MEGABLAST are word
  length (min. exact match), ID cutoff, and filter
  disabling (will ignore repeats instead of marking
  them)
• Also useful for cross-genome comparisons

17
DNA Folding

• The most commonly used tool for check the folding
  confirmation of nucleice acids is the
  DNA-FOLD/RNA-FOLD family of programs from Zuker.
  We use this to check hairpins, hopefully no more
  than 10 to 13kcal/mol.
• A member of the package useful for large scale
  analysis (many small seqs) is quikfold sic,
  which produces only thermodynamic statistics
  rather than the pretty fold images available from
  the other programs

18
Secondary Binding

• The dataset to check against for a microarray
  would be the all the ORFs (prokaryotes) or all of
  the known cDNA (eukaryotes)
• For sequencing it would be the vector and any
  known sequence from the clone
• Most software uses a ID cutoff, but the location
  of the mismatches can greatly change the effect
  on melting temperature!

19
Accurate Secondary Binding Checks

• Free energy (G) is derived from the same enthalpy
  and entropy values as temperature, so why not use
  a dynamic method to optimize it in sequence
  searches? ?Gº ?Hº - T?Sº

Free energy (G) vs. Tm for left to right, random
50-, 25-, 20-, 15-, and 10-mers
20
Profile Alignments

• Hidden Markov Models such as found in the Pfam
  database, and Profiles such as found in ProSite
  are types of Position Specific Scoring Matrices
  (PSSMs).
• At each model position, a score is given to each
  possible match/mismatch. An A-G mismatch in
  different positions may score differently,
  usually based on frequencies in MSAs. Below is
  the start of a Pfam protein model match

21
Profile Alignments

• We can also use the PSSM to encode nearest
  neighbour free energy (G) thermodynamics
• Note that the first A matches with a score of
  580, the second with a score of 1000, this is a
  position specific differences because they have
  different neighbours (TA vs. AA)

22
Caveats

• Profiles, like Smith-Waterman alignments, use
  dynamic methods, therefore the optimal solution
  will always be found
• Using BLAST to find secondary binding, you may
  miss matches with good thermodynamics, because of
  the shortcuts used to make BLAST fast.
• At the very least use a dynamic method to check
  oligo secondary matches. Profiles will
  additionally tell you not only a ID, but the
  free energy

23
Results Interpretation

• Given a free energy cutoff, we can check all
  secondary results for their melting temperature,
  and ensure it is outside the margin desired by
  the user (e.g. more than 10 degrees below the
  melting temperature of the target duplex)

24
Thermodynamics Encoding Rules

• Using a profile we can incorporate the most
  important thermodynamic values at each position
• 1. a match score is the molar caloric free energy
  contribution of the matched base and its 5
  neighbor, and a portion of the unified models
  length-dependent salt concentration penalty.

25
Rules cont'd

• 2. a mismatch score include the free energy
  contribution of (a) the matched 5 neighbor and
  mismatched base (b) the matched 5 neighbor and
  mismatched base on the opposite strand (c)
  discount for the NN contribution in the next
  position

26
Rules cont'd

• 3. the gap insertion penalty reflects the NN free
  energy penalty for single base bulges in a duplex
• 4. the start of the sequence encodes the unified
  models self-complementarity penalties if
  applicable. Mismatches in this position also
  encode mismatched end thermodynamics.

27
Thermodynamic Encoding

• 5. one extra state at each end of the profile
  sequence encodes dangling end thermodynamics in
  the case that the oligo matches to either
  terminus.
• These rules are far too complicated to encode in
  a standard pairwise scoring matrix, therefore we
  use PSSMs.
• We use Profiles rather than HMMs because the
  scores in HMMs must sum to 1 after a conversion
  process, a restriction we can't comply with.

28
Hardware Acceleration

• Dynamic methods such as PSSMs are expensive, but
  we have hardware accelerators.
• Decypher runs these searches at least 50 times
  faster than software, so we can use dynamic
  methods on the Sulfolobus genome and design a
  chip of 3456 optimized oligos in 1.5 hours

29
Data Parallelism

• In addition to using hardware accelerators, we
  can use software parallelization to speed up
  processing, since the calculation of any oligo is
  independent from the calculation of any other.
  You can apply this prinicipal generally in your
  analyses break down your query files into chunks
  that can be run in different shells, then
  concatenated at the end.

30
Web Interface

• A Web interface using the hardware acceleration
  for the design of microarrays is available at
• http//osprey.ucalgary.ca/
• Your assignment
• Separately try some of the component tools
  Osprey uses
• Try the Osprey interface