The document discusses biological computation and gene regulation in cells. It describes how (1) cells perform biochemical information processing to transform cues into biological functions, (2) embryonic stem cells can adopt different states through gene regulation, and (3) techniques like Boolean networks and satisfiability modulo theories can be used to model and analyze gene interaction networks inferred from experimental data. The techniques allow predicting cell behaviors and identifying gene interaction programs governing processes like stem cell differentiation.
21. Modelling Dynamic Gene Interactions
Boolean Network State transition system
111 110
011 010
101
001
100000
0 1 1
A B C
Bit-vector
Kauffman, 1969
AND
OR
23. Challenge: In reality, experimental data only indicates
the possibility of an interaction
Abstract Boolean
Network
(ABN)
Classifying π interactions as possible defines 2 π unique topologies
24. Challenge: Unique networks can produce the same
behaviour
111 110
011 010
100001
101
000
110 111
011
001101
010
100
000
We need to consider the set of potential models
27. βIf gene A is active, then
gene B is active or gene
C is active.β
π΄ β (π΅ β¨ πΆ)
βIn Experiment 1, initially A
is inactive and at step 6 it
is active.β
πΈ0
1
. π΄ = 0 β§ πΈ6
1
. π΄ = 1
Gene
A
Gene
B
Gene
C
28. Boolean Satisfiability (SAT)
1. Given a propositional formula π, determine if there is a variable assignment such that π evaluates to TRUE.
2. Generate a model that satisfies π.
π β π΄ β (π΅ β¨ πΆ)
βIf gene A is active, then gene
B is active or gene C is active.β
{π΄, π΅, πΆ}
{Β¬π΄, Β¬π΅, Β¬πΆ}
{π΄, Β¬π΅, πΆ}
Assignments under which π is TRUE
{π΄, π΅, Β¬πΆ} π is SATISFIABLE
29. Satisfiability Modulo Theories (SMT)
β’ Decision procedures for pre-defined theories
β’ Boolean
β’ Uninterpreted functions
β’ Integers
β’ Bit-vectors
β’ Floating point numbers
β’ Data types (Strings, Arrays, Lists, etc.)
β’ Theory combination strategy
β’ Standardisation
39. Identifying Possible Gene Interactions
Pearson
coefficient:
-0.98
Pearson
coefficient:
0.98
Experimental data describing the expression of
key genes under different inputs
40. Gene expression correlation does not indicate which
gene behaves as the regulator
Abstract Network Topology
Four unique possibilities
41. ES cells can efficiently convert between
culture conditions
Experimental Constraints
Transform into constraints on network
trajectories
50. β’ Uncovered the biological program governing stem
cell decision-making
β’ Consistent with 149 different experiments
β’ Predictive accuracy of 80%
β’ Used to inform biological experiments
β’ Increased reprogramming efficiency to 100%, and
reduced time to just 24hrs
β’ Predicted gene activation trajectories substantiated
even at single cell resolution
Dunn & Li et al., EMBO J (2019)
51.
52.
53.
54. Acknowledgements
β’ Amy Li, Graziano Martello and Austin Smith
β’ Boyan Yordanov, Hillel Kugler, Christoph Wintersteiger
β’ www.research.microsoft.com/rein
55. Paul
Grant
Neil
Dalchau
Boyan
Yordanov
Carlo
Spaccasassi
Collaborators
Programming Stem Cells
University of Cambridge Stem Cell Institute: Austin Smith
University of Cambridge Metabolic Research Laboratories: Davide
Chiarugi, Anne-Claire Guenantin, Antonio Vidal-Puig
University of Padova: Graziano Martello
Programming DNA
Princeton University: Bonnie Bassler
University of Washington: Georg Seelig, Gourab Chatterjee, Suzie Pun
University of New Mexico: Matthew Lakin
Rice University: Dave Zhang
University of Cambridge: Ulrich Keyser, Elisa Hemmig
Microsoft Research: Karin Strauss, Yuan Chen
Caltech: Frits Dannenberg
Programming Genetic Devices
University of Cambridge: Jim Ajioka, Jim Haseloff, Om Patange,
Eugene Nadezhdin
UCL: Chris Barnes, Luca Rosa
Luca
Cardelli
Filippo
Polo
Colin
Gravill
Sara-Jane
Dunn
James
Locke
Andrew
Phillips
Jacob
Halatek
Prashant
Vaidyanathan