The document discusses a project to build better drug discovery software by engaging scientists. It involves:
1) Understanding the complex workflows of pooled library screening through user research with scientists.
2) Designing a suite of web applications called ABACUS to integrate the entire screening process and store data, with participation from scientists and stakeholders.
3) Testing the solution through an agile delivery model with scientist feedback, focusing on functionality before aesthetics.
3. NIBR NX
Cancer Genetic mutation Therapy
Chronic Myelo. Leukemia BCR-ABL translocation Gleevec
Melanoma BRAF mutation Dabrafenib
PTC
Cancer is a Genetic Disease
Rationale for Targeted Therapy - Find the oncogene & hit it with a drug
3
4. NIBR NX
Pooled library screening
Business Use Only4
Lets you ask the cancer cells what they need to live
Lentiviral shRNA pool
Add to Cells Select
Initial representation Final representation
Culture
10-14
Days
deep sequencing
WT Mutant
Growth
promotion
Growth
inhibition
shRNAs
Candidates for
Targeted Therapies
Cell lines
5. NIBR NX
Pooled Screen Workflow overview
Complex multistep process with many different data and user types
Library Building
• Design
• Cloning
• Virus Packaging
Cell Culture
• Infect with libraries
• Add treatments
• Collect cell pellets
gDNA prep and PCR/NGS
• gDNA extraction
• PCR automation
• NGS
Data Analysis
ABACUS
6. NIBR NX
ABACUS: Suite of web applications
6
Ties together entire pooled screen workflow and stores the data
Design Experiments
Track samples
Store sequencing results
Group and analyze data
Register samples
7. NIBR NX
Analysis of pooled library screening
7
Informatics and Outlier Metrics
(A) screening and analytical workflow. 2-week pooled shRNA viability
screens were followed by an NGS readout and shRNA gene-level
aggregation by RSA and ATARiS. Feature correlation was performed
using k-means clustering (k = 3) to identify sensitive and insensitive
populations with resulting display of top sensitivity correlations. KRAS
example is shown.
(B) gene activity categorization into inert, active, or essential profiles.
Essential gene dependency correlations include self-CN/expr and AGO2
CN/expr correlations. RSA waterfall plot for PHF5A, colored by PHF5A
copy number. RSA sensitivity cutoff shown at −3 indicates that this is an
essential gene in most lines tested. PSMC5 RSA sensitivity plotted against
AGO2 CN and colored by AGO2 expression.
(C) Normality LRT compares the fit of a skewed Student’s t distribution and
a normal distribution. Profiles with better skewed Student’s t distribution fit
over normal have high NormLRT scores.
(D) Top outliers by class colored by their presence in COSMIC. Non-
COSMIC genes labeled and TFs shown as triangles.
Cell 2017 170, 577-592.e10DOI: (10.1016/j.cell.2017.07.005)
10. NIBR NX
Do not get distracted by
the complexity of the
science.
Business Use Only10
Key
Learning
11. NIBR NX
About the Project
• Full commitment from Science and IT
• Clear goals and deliverables
• User Centered Design
• Agile delivery model
• The Team
– Delivery Lead/Project Manager/Scrum Master
– UX Researcher/Designer
– Business Analyst
– Software Engineers
– Data Engineer
– Multiple scientists engaged throughout
Business Use Only11
36. NIBR NX
Scientist Perspective – What worked
well
• Involve scientists early and often
• Be able to teach and learn on both sides
– Thorough explanation of our process was very helpful
• Got what we needed, not what we asked for
• All stakeholders making decisions together, resolving
differences together
– Suite of tools, not one large tool
• Leadership buy-in on both sides is key
Business Use Only36
37. NIBR NX
Scientist Perspective – What was
difficult
• Working style and time expectations are very different
• Finding a common language, two ways
• Reassuring leadership that time investment is being
productive prior to release
– Lots of time/work before you see a product
Business Use Only37