280 likes | 429 Views
Boolean Networks and Experiment Design B-Cell Single Ligand Screen. Stuart Johnson Bioinformatics and Data Analysis Lab UCSD. Outline. Why Boolean networks? Building/Displaying Boolean Networks Experiment design Procedure Some competing (sub)networks from the B-Cell data Conclusions.
E N D
Boolean Networks and Experiment DesignB-Cell Single Ligand Screen Stuart Johnson Bioinformatics and Data Analysis Lab UCSD
Outline • Why Boolean networks? • Building/Displaying Boolean Networks • Experiment design • Procedure • Some competing (sub)networks from the B-Cell data • Conclusions
Why try Boolean Networks? • Model • Biochemical system • lots of complexity • predictive • lots of meaning very difficultinverse problem doable forward problem • Data • noisy • partial sampling
Why try Boolean Networks? • Boolean networks • some complexity • predictive (exp. design) • data-like • meaning? consistency = causality; should tell us about connectivity easy Boolean data easy
2nd msg / co-sampled Ca Boolean data TIME P-P, 2nd Msg red=1 at 99% confidence: P(d=NC)<.01 blue=0 everythingelse Experimental conditions
Phosphoproteins Boolean data TIME P-P, 2nd Msg red=1 at 99% confidence: P(d=NC)<.01 blue=0 everythingelse Experimental conditions
late resp. Ca -> PP Ca,cAMP -> No PP early resp. groups ofsiml. resp. Boolean data TIME P-P, 2nd Msg Experimental conditions
Boolean data TIME P-P, 2nd Msg Node = Full column of data; all exp. cond. Experimental conditions
Gq Known ligand/ receptor interactions from AfCS ligand descriptions Inputs, etc. Experimental Conditions Single ligand screen inputs
consistent? ? Extractingpatterns AIG Ca (.5 min) ELC LPA Experimental Conditions
LPA0 Ca0.5 ER12.5 ERK1 (2.5 min) 0 0 0 1 0 ? 0 1 1 1 1 0 Time Experimental Conditions Graph TruthTable displayingand encodingpatterns
all hypotheses:ER1(2.5 min) H1 H2 H1,H2 & H3: Early calcium is associated with ER1 H1: LPA is special (causes an early Ca signal but no ER1) H2: M3A is special (0.5 min Ca, no 1 min Ca, but ER1) H3: no special ligands, ER1 consistent with Ca & cAMP H3
Constructing complete networks Input nodes I1 I2 I3 nodes with truth tables N1 N2 N3 5 x 7 x 3 = 105networksmaximum
Constructing complete networks I1 I2 I3 N1 N2 N3
Constructing complete networks I1 I2 I3 N1 N2 N3
Constructing complete networks I1 I2 I3 N1 N2 N3 • “Feedback” not allowed! a completely determined network can have multiple output states; forward and inverse problems no longer “easy”
1 output state Experiment Design: networks reproduceresults of completed experiments • All networks: 1 possible output state: • For known inputs, every network simply reproduces results of completed experiments • (Information) Entropy = score = 0
3 output states Experiment Design: networks are predictive • All networks: multiple possible output states: • these multiple states correspond to unknown entries (?) in truth tables and the different connectivity of the networks • Entropy = score > 0
Dual-ligand experiment design ligand 2 ligand 1 entropy score
Dual-ligand experiment design ligand 2 ligand 1 entropy score ELC + LPA
Procedure Do Experiments DisplayBooleanNetworks BuildBoolean Networks Score classof experiments pick highest scoring exp.
LIG LIG LIG LIG 2M RCP 2M RCP 2M 2M PP PP 1 PP PP 1 PP PP Controlling Complexity: Constraint Graphs • Graphs specify allowable inputs and hops
LIG LIG LIG LIG 2M RCP 2M RCP 2M 2M PP PP 1 PP PP 1 PP PP Controlling Complexity: Constraint Graphs • Graphs specify allowable inputs and hops
LIG RCP 2M PP 1 PP Network display All node rules • Can filter/cluster/display these rules to see: • ligand classification (chemokines, cytokines, etc) • clusters of similar control patterns • etc. - “pathways”
LIG 2M PP Early Calciumvs ... Early Calcium+ cAMP
LIG 2M PP 1 PP ER1 -> ER2,P90 P90 -> AKT ST6 -> ST3
LIG RCP 2M PP Early Ca & Gqcontrol vs ... Early Ca& G12
Conclusions • This is a general method/implementation and will extend to the RAW screens and FXM in some form • Boolean network analysis has many interesting features: • learns from experiments/proposes new exp. • formalizes inclusion of known information as either constraint graphs or hidden nodes • caveat 1: the BN have encoded any real meaning • caveat 2: you can control complexity and digest the networks inferred • http://dev.afcs.org:12057/ for the latest results, navigable/clickable networks and more background