Small Molecules and siRNA: Methods to Explore Bioactivity Data
Upcoming SlideShare
Loading in...5
×

Like this? Share it with your network

Share
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
    Be the first to like this
No Downloads

Views

Total Views
875
On Slideshare
875
From Embeds
0
Number of Embeds
0

Actions

Shares
Downloads
13
Comments
0
Likes
0

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide
  • Outliers in a cliff prediction model are not as severe since SALI changes more slowly than just activity differences
  • For SALI = 0, had to set log10(SALI) = 0Similar performance if we use SALI and not log10(SALI) at least more % variance is explained. Still fail on most significant cliffs
  • View plates (raw, normalized, adjusted, …)Highlight specific genes, siRNA’sView assay statisticsView pathway membership (via Wikipathways)Linkout to external resources (Entrez, GeneCards, …)Hit selection, follow up (DRC)
  • View plates (raw, normalized, adjusted, …)Highlight specific genes, siRNA’sView assay statisticsView pathway membership (via Wikipathways)Linkout to external resources (Entrez, GeneCards, …)Hit selection, follow up (DRC)
  • * Proscillaridin A was not selected in the 20 compounds for further analysis in the paper* 2 cardiac glycosides in the top 3, target appears to be caspase-3 (activating it). CG inhibition of NF-kb is well known . See PNAS 2005, by Pollard* Trabectidin induces lethal DNA strand breaks and blocks cell cycle in G2 phase
  • PSM* genes code for proteosome subunits – so they likely prevent the ubiquination of the IkBa complex, so that RelA+cp50 cannot be released from the IkBa complex and enter the nucleus
  • Size of node indicates potency – larger is more potentLanatosidec and a have Tc = 1 and hence the edge was not shown (ideally it should be shown)
  • Good confirmation that SEA worksSize of node corresponds to SEA confidence score
  • We consider 41 compounds rather than 55, since a number of them did not have sufficiently confident target predictionsWe then get to 18 compounds since, many of the predicted genes, did not map to an NCI PID pathway
  • Pheontypic difference can arise when PPI’s are involved
  • HPRD subnetwork corresponding to the Qiagen HDG has 6782 genes
  • HPRD subnetwork corresponding to the Qiagen HDG has 6782 genes

Transcript

  • 1. Small Molecules and siRNA:Methods to Explore Bioactivity Data
    Rajarshi Guha
    NIH Chemical for Translational Therapeutics
    August 17, 2011
    Pfizer, Groton
  • 2. Background
    Cheminformatics methods
    QSAR, diversity analysis, virtual screening, fragments, polypharmacology, networks
    More recently
    siRNAscreening, high content imaging,combination screening
    Extensive use of machine learning
    All tied together with software development
    Integrate small molecule information & biosystems – systems chemical biology
  • 3. Outline
    Exploring the SAR landscape
    The landscape view of SAR data
    Quantifying SAR landscapes
    Extending an SAR landscape
    Linking small molecule & RNAiHTS
    Overview of the Trans NIH RNAi Screening Initiative
    Infrastructure components
    Linking small molecule & siRNA screens
  • 4. The Landscape View of Structure Activity Datasets
  • 5. Structure Activity Relationships
    Similar molecules will have similar activities
    Small changes in structure will lead to small changes in activity
    One implication is that SAR’s are additive
    This is the basis for QSAR modeling
    Martin, Y.C. et al., J. Med. Chem., 2002, 45, 4350–4358
  • 6. Structure Activity Landscapes
    Rugged gorges or rolling hills?
    Small structural changes associated with large activity changes represent steep slopes in the landscape
    But traditionally, QSAR assumes gentle slopes
    Machine learning is not very good for special cases
    Maggiora, G.M., J. Chem. Inf. Model., 2006, 46, 1535–1535
  • 7. Characterizing the Landscape
    A cliff can be numerically characterized
    Structure Activity Landscape Index (SALI)
    Cliffs are characterized by elements of the matrix with very large values
    Guha, R.; Van Drie, J.H., J. Chem. Inf. Model., 2008, 48, 646–658
  • 8. Visualizing SALI Values
    The SALI graph
    Compounds are nodes
    Nodes i,j are connected if SALI(i,j) > X
    Only display connected nodes
  • 9. What Can We Do With SALI’s?
    SALI characterizes cliffs & non-cliffs
    For a given molecular representation, SALI’s gives us an idea of thesmoothness of the SAR landscape
    Models try and encodethis landscape
    Use the landscape to guidedescriptor or model selection
  • 10. Descriptor Space Smoothness
    Edge count of the SALI graph for varying cutoffs
    Measures smoothness of the descriptor space
    Can reduce this to a single number (AUC)
  • 11. Other Examples
    Instead of fingerprints, we use molecular descriptors
    SALI denominator now uses Euclidean distance
    2D & 3D random descriptor sets
    None are really good
    Too rough, or
    Too flat
    2D
    3D
  • 12. Feature Selection Using SALI
    Surprisingly, exhaustive search of 66,000 4-descriptor combinations did not yield semi-smoothly decreasing curves
    Not entirely clear what type of curve is desirable
  • 13. Measuring Model Quality
    A QSAR model should easily encode the “rolling hills”
    A good model captures the most significantcliffs
    Can be formalized as
    How many of the edge orderings of a SALI graph does the model predict correctly?
    Define S (X ), representing the number of edges correctly predicted for a SALI network at a threshold X
    Repeat for varying X and obtain the SALI curve
  • 14. SALI Curves
  • 15. Model Search Using the SCI
    We’ve used the SALI to retrospectively analyze models
    Can we use SALI to develop models?
    Identify a model that captures the cliffs
    Tricky
    Cliffs are fundamentally outliers
    Optimizing for good SALI values implies overfitting
    Need to trade-off between SALI & generalizability
  • 16. Predicting the Landscape
    Rather than predicting activity directly, we can try to predict the SAR landscape
    Implies that we attempt to directly predict cliffs
    Observations are now pairs of molecules
    A more complex problem
    Choice of features is trickier
    Still face the problem of cliffs as outliers
    Somewhat similar to predicting activity differences
    Scheiber et al, Statistical Analysis and Data Mining, 2009, 2, 115-122
  • 17. Motivation
    Predicting activity cliffs corresponds to extending the SAR landscape
    Identify whether a new molecule will perform better or worse compared to the specific molecules in the dataset
    Can be useful for guiding lead optimization, but not necessarily useful for lead hopping
  • 18. Predicting Cliffs
    Dependent variable are pairwise SALI values, calculated using fingerprints
    Independent variables are molecular descriptors – but considered pairwise
    Absolute difference of descriptor pairs, or
    Geometric mean of descriptor pairs

    Develop a model to correlate pairwise descriptors to pairwise SALI values
  • 19. A Test Case
    We first consider the CavalliCoMFA dataset of 30 molecules with pIC50’s
    Evaluate topological and physicochemical descriptors
    Developed random forest models
    On the original observed values (30 obs)
    On the SALI values (435 observations)
    Cavalli, A. et al, J Med Chem, 2002, 45, 3844-3853
  • 20. Double Counting Structures?
    The dependent and independent variables both encode structure.
    But pretty low correlations between individual pairwisedescriptors and the SALI values
  • 21. Model Summaries
    Original pIC50
    RMSE = 0.97
    SALI, AbsDiff
    RMSE = 1.10
    SALI, GeoMean
    RMSE = 1.04
    All models explain similar % of variance of their respective datasets
    Using geometric mean as the descriptor aggregation function seems to perform best
    SALI models are more robust due to larger size of the dataset
  • 22. Test Case 2
    Considered the Holloway docking dataset, 32 molecules with pIC50’s and Einter
    Similar strategy as before
    Need to transform SALI values
    Descriptors show minimal correlation
    Holloway, M.K. et al, J Med Chem, 1995, 38, 305-317
  • 23. Model Summaries
    Original pIC50
    RMSE = 1.05
    SALI, AbsDiff
    RMSE = 0.48
    SALI, GeoMean
    RMSE = 0.48
    The SALI models perform much poorer in terms of % of variance explained
    Descriptor aggregation method does not seem to have much effect
    The SALI models appear to perform decently on the cliffs – but misses the most significant
  • 24. Model Summaries
    Original pIC50
    RMSE = 1.05
    SALI, AbsDiff
    RMSE = 9.76
    SALI, GeoMean
    RMSE = 10.01
    With untransformed SALI values, models perform similarly in terms of % of variance explained
    The most significant cliffs correspond to stereoisomers
  • 25. Test Case 3
    38 adenosine receptor antagonists with reported Ki values; use 35 for training and 3 for testing
    Random forest model on the SALI values performed reasonable well (RMSE = 7.51, R2=0.62)
    Upper end ofSALI rangeis better predicted
    Kalla, R.V. et al, J. Med. Chem., 2006, 48, 1984-2008
  • 26. Test Case 3
    • The dataset does not containing really big cliffs
    • 27. Generally, performance is poorer for smaller cliffs
    For any given hold out molecule, range of error in SALI prediction is large
    Suggests that some form of domain applicability metric would be useful
  • 28. Model Caveats
    Models based on SALI values are dependent on their being an SAR in the original activity data
    Scrambling results for these models are poorer than the original models but aren’t as random as expected
  • 29. Conclusions
    SALI is the first step in characterizing the SAR landscape
    Allows us to directly analyze the landscape, as opposed to individual molecules
    Being able to predict the landscape could serve as a useful way to extend an SAR landscape
  • 30. Joining the Dots: Integrating High Throughput Small Molecule and RNAi Screens
  • 31. RNAi Facility Mission
    Pathway (Reporter assays, e.g. luciferase, b-lactamase)
    Simple Phenotypes (Viability, cytotoxicity, oxidative stress, etc)
    Perform collaborative genome-wide RNAi screening-based projects with intramural investigators
    Advance the science of RNAi and miRNA screening and informatics via technology development to improve efficiency, reliability, and costs.
    Complex Phenotypes (High-content imaging, cell cycle, translocation, etc)
    Range of Assays
  • 32. RNAi Informatics Infrastructure
  • 33. RNAi Analysis Workflow
    Raw and Processed Data
    GO annotations
    Pathways
    Interactions
    Hit List
    Follow-up
  • 34. RNAi Informatics Toolset
    Local databases (screen data, pathways, interactions, etc).
    Commercial pathway tools.
    Custom software for loading, analysis and visualization.
  • 35. Back End Services
    Currently all computational analysis performed on the backend
    R & Bioconductor code
    Custom R package (ncgcrnai) to support NCGC infrastructure
    Partly derived from cellHTS2
    Supports QC metrics, normalization, adjustments, selections, triage, (static) visualization, reports
    Some Java tools for
    Data loading
    Library and plate registration
  • 36. User Accessible Tools
  • 37. User Accessible Tools
  • 38. RNAi& Small Molecule Screens
    CAGCATGAGTACTACAGGCCA
    TACGGGAACTACCATAATTTA
    What targets mediate activity of siRNA and compound
    Pathway elucidation, identification of interactions
    • Reuse pre-existing MLI data
    • 39. Develop new annotated libraries
    Target ID and validation
    Link RNAi generated pathway peturbations to small molecule activities. Could provide insight into polypharmacology
    • Run parallel RNAi screen
    Goal: Develop systems level view of small molecule activity
  • 40. HTS for NF-κB Antagonists
    NF-κB controls DNA transcription
    Involved in cellular responses to stimuli
    Immune response, memory formation
    Inflammation, cancer, auto-immune diseases
    http://www.genego.com
  • 41. HTS for NF-κB Antagonists
    ME-180 cell line
    Stimulate cells using TNF, leading to NF-κB activation, readout via a β-lactamase reporter
    Identify small molecules and siRNA’s that block the resultant activation
  • 42. Small Molecule HTS Summary
    2,899 FDA-approved compounds screened
    55 compounds retested active
    Which components of the NF-κB pathway do they hit?
    17 molecules have target/pathway information in GeneGO
    Literature searches list a few more
    Most Potent Actives
    Proscillaridin A
    Trabectidin
    Digoxin
    Miller, S.C. et al, Biochem. Pharmacol., 2010, ASAP
  • 43. RNAi HTS Summary
    Qiagen HDG library – 6886 genes, 4 siRNA’s per gene
    A total of 567 genes were knockeddown by 1 or more siRNA’s
    We consider >= 2 as a “reliable” hit
    16 reliable hits
    Added in 66 genes for follow up via triage procedure
  • 44. The Obvious Conclusion
    The active compounds target the 16 hits (at least) from the RNAi screen
    Useful if the RNAi screen was small & focused
    But what if we’re investigating a larger system?
    Is there a way to get more specific?
    Can compound data suggest RNAi non-hits?
  • 45. Small Molecule Targets
    Bortezomib (proteosome inhibitor)
    Some small molecules interact with core components
    Daunorubicin (IκBα inhibitor)
  • 46. Small Molecule Targets
    Montelukast (LDT4 antagonist)
    Others are active against upstream targets
    We also get an idea of off -target effects
  • 47. Compound Networks - Similarity
    Evaluate fingerprint-based similarity matrix for the 55 actives
    Connect pairs that exhibit Tc> 0.7
    Edges are weightedby the Tc value
    Most groupings areobvious
  • 48. A “Dictionary” Based Approach
    Create a small-ish annotated library
    “Seed” compounds
    Use it in parallel small molecule/RNAi screens
    Use a similarity based approach to prioritize larger collections, in terms of anticipated targets
    Currently, we’d use structural similarity
    Diversity of prioritized structures is dependent on the diversity of the annotated library
  • 49. Compound Networks - Targets
    Predict targets for the actives using SEA
    Target based compound network maps nearly identically to the similarity based network
    But depending on the predicted target qualitywe get poor (or no) mappings to the RNAi targeted genes
    Keiser, M.J. et al, Nat. Biotech., 2007, 25, 197-206
  • 50. Gene Networks - Pathways
    Nodes are 1374 HDG genes contained in the NCI PID
    Edge indicates two genes/proteins are involved in the same pathway
    “Good” hits tend to be very highly connected
    Wang, L. et al, BMC Genomics, 2009, 10, 220
  • 51. (Reduced) Gene Networks – Pathways
    Nodes are 526 genes with >= 1 siRNA showing knockdown
    Edge indicates two genes/proteins are involved in the same pathway
  • 52. Pathway Based Integration
    Direct matching of targets is not very useful
    Try and map compounds to siRNA targets if the compounds’ predicted target(s) and siRNA targets are in the same pathway
    Considering 16 reliable hits, we cover 26 pathways
    Predicted compound targets cover 131 pathways
    For 18 out of 41 compounds
    3 RNAi-derived pathways not covered by compound-derived pathways
    Rhodopsin, alternative NFkB, FAS
  • 53. Pathway Based Integration
    Still not completely useful, as it only handled 18 compounds
    Depending on target predictions is probably not a great idea
  • 54. Integration Caveats
    Biggest bottleneck is lack of resolution
    Currently, both small molecule and RNAi data are 1-D
    Active or inactive, high/low signal
    CRC’s for small molecules alleviate this a bit
    High content screens can provide significantly more information and so better resolution
    Data size & feature selection are of concern
  • 55. Integration Caveats
    Compound annotations are key
    Currently working on using ChEMBL data to provide target ‘suggestions’
    More comprehensive pathway data will be required
    RNAi and small molecule inhibition do not always lead to the same phenotype
    Could be indicative of promiscuity
    Could indicate true biological differences
    Weiss, W.A. et al, Nat. Chem. Biol., 2007, 12, 739-744
  • 56. Conclusions
    Building up a wealth of small molecule and RNAi data
    “Standard” analysis of RNAi screens relatively straightforward
    Challenges involve integrating RNAi data with other sources
    Primary bottleneck is dimensionality of the data
    Simple flourescence-based approaches do not provide sufficient resolution
    High-content is required
  • 57. Acknowledgements
    John Van Drie
    Gerry Maggiora
    MicLajiness
    JurgenBajorath
    Scott Martin
    Pinar Tuzmen
    CarleenKlump
    DacTrung Nguyen
    Ruili Huang
    Yuhong Wang
  • 58. CPT Sensitization & “Central” Genes
    Yves Pommier, Nat. Rev. Cancer, 2006.
    TOP1 poisons prevent DNA religation resulting in replication-dependent double strand breaks. Cell activates DNA damage response (e.g. ATR).
  • 59. Screening Protocol
    Screen conducted in the human breast cancer cell line MDA-MB-231. Many variables to optimize including transfection conditions, cell seeding density, assay conditions, and the selection of positive and negative controls.
  • 60. Hit Selection
    Follow-Up Dose Response Analysis
    ATR
    Screen #1
    siNeg
    siATR-A
    siATR-B
    siATR-C
    Viability (%)
    Sensitization Ranked by Log2 Fold Change
    CPT (Log M)
    Screen #2
    MAP3K7IP2
    siNeg
    siMAP3K7IP2-A
    siMAP3K7IP2-B
    siMAP3K7IP2-C
    Viability (%)
    siMAP3K7IP2-D
    Sensitization Ranked by Log2 Fold Change
    CPT (Log M)
    Multiple active siRNAs for ATR, MAP3K7IP2, and BCL2L1.
  • 61. Are These Genes Relevant?
    Some are well known to be CPT-sensitizers
    Consider a HPRD PPI sub-network corresponding to the Qiagen HDG gene set
    How “central” are these selected genes?
    Larger values of betweennessindicate that the node lies onmany shortest paths
    Makes sense - a number of them are stress-related
    But some of them have very lowbetweenness values
  • 62. Are These Genes Relevant?
    Most selected genesare densely connected
    A few are not
    Generally did notreconfirm
    Network metrics could be used to provide confidencein selections