1. IPANEMAP Integrative Probing Analysis of Nucleic Acids Empowered by Multiple Accessibility Profiles
Afaf SAAIDI & Yann PONTY Bruno SARGUEIL & Delphine ALLOUCHE
CNRS/Ecole Polytechnique Univ. Paris Descartes

2. Structure probing in a nutshell

3. SHAPE + 2D predictions, still below 110%…
RSample, Spasic et al, NAR 2018
David H Mathews

4. The practice of RNA modeling
Probing can be tricky to
perform and interpret.
Modelers typically use combinations of techniques and reagents, but integration is no gimme…
Bruno Sargueil
How to integrate multiple probing profiles?

5. A KISS approach
Native structure should be represented in each of the pseudo-Boltzmann ensembles (maybe not as MFE)
Reactivity profiles uniformly captured as pseudo potentials (aka soft constraints) during sampling (RNAsubopt)
ACGAUGAUCGACUACGAUCGA
UCGACUAGCUACGUACUGACU
CGGCUAGAUUAGCUUAUGA…
Sampling (RNAsubopt)
Pseudo-Boltzmann

6. IPANEMAP Pipeline

7. Single condition/single structure dataset
IPANEMAP vs Rsample
[Spasic et al, NAR 2018]
Hajdin et al dataset SHAPE 1M nts RNAs
80% GM IPANEMAP vs 63% GM for Rsample
Reason: Rsample ≈ MEA
Geometric Mean

8. Multiple conditions improve predictions?
6 RMDB RNAs
5s RNA E. coli
Glycine Riboswitch F. Nucleatum
cidGMP riboswitch V. Cholerae
P4 - P6 domain Tetrahymena ribozyme
add Adenine Riboswitch
tRNA phenylalanine yeast
3 conditions
SHAPE (NMIA)
DMS
CMCT

9. Back to the lab for even more data
Lariat Capping Ribozyme (PDB 4P8Z) from Didymium iridis
Probed under 16 different conditions:
It has a pseudoknot
Technique: SHAPE, DMS, CMCT
Reagents: 1M7, NMIA, DMS, CMCT, NAI, BzCN
Stop based and Mutation based readouts
Presence/absence Magnesium

10. Mono probing analysis IPANEMAP ≥ MFE, MEA Mutations > Stops
+Mg2+ > -Mg2+
Clear outliers
MCC:

11. Mono probing analysis 16 conditions + no-probing (-)
(Squared) Euclidian distance between conditions
16 conditions + no-probing (-)
Conditions cluster in 8 groups
2 singletons
Outliers confirmed

12. Dual probing analysis
Pairs of conditions do not significantly improve predictions but:
Mitigate the risk of poor conditions: % MCC on average against worst condition
Slightly increase expectation: % avg (+2% median) against mean MCC, +4.4% w/o outliers
Sensitive to contamination: % MCC against best, -3.3% w/o outliers

13. Triplets of conditions improve predictions
Considering triplets of conditions (560 triplets) :
Improves by 9.1% against worst of triplet
Improves by 1.7% vs mean MCC (+6.9% w/o outliers)
Some triplets improve overall best MCCs
1M7I-MAPMGIL + 1M7-MAPIL + DMS-CEMG
1M7I-MAPMGIL + 1M7-MAPIL + NMIA-CEMG
1M7I-MAPMGIL + DMS-CEMG + CMCT-CEMG
1M7I-MAPMGIL + 1M7I-MAPIL-3D + BzCN-CEMG
Beyond three conditions?
→ 85.3%MCC

14. The full Monty: Multi probing
All 16 conditions together: 80% MCC (vs 70.5% mono Avg)
Within clustered conditions: No info/improvement
Across clustered conditions (One condition per cluster)
77% MCC w/o NAI (+6.5)
72% MCC with NAI (+1.5)
Orthogonality?
Complementarity?

15. Treating mutants as conditions
Mutate-And-Map profiles produced by Das lab (RMDB)
Systematic single-point mutants, usually similar structure

16. Conclusion https://github.com/afafbioinfo/IPANEMAP
IPANEMAP is way more than just a pretty acronym
Exploits the complementarity of probing experiments
Paves the way for quantifying a notion of orthogonality
Take home message:
Probing remains challenging (accuracy still slightly below 120%...)
Having >2 reactivity profiles (even bad ones) appears to help
The future:
Better pseudo-potentials & clustering procedures
Test on multi-stable RNAs
Optimal/iterative design of probing experiments

17. Thank you! Ronny Lorenz for modular extensions to Vienna package
Afaf Saaidi
Bruno Sargueil
Delphine Allouche
GaTech/UNC
Paris Descartes
Ronny Lorenz for modular extensions to Vienna package
Soft constraints x Sampling = Awesomeness
Supported by Fondation pour la Recherche Médicale

18. Reproducibility

19. 2D folding prediction paradigms
Probability/Concentration
0.5 1 B C D A C A B C D D
Time A B C D B
Enzymatic degradation
RNA half life A B C D A B C D A
Equilibrium A B C D
MFE