A logic modelling workflow for systems pharmacology Luis Tobalina 13/07/2018 www.saezlab.org Institute for Computational Biomedicine @sysbiomed Heidelberg University & RWTH Aachen
Outline • Context • Pipeline • Network • Modelling • Data • Insights • Summary � 2
Systems Biomedicine • Systems Biology examines how cell components interact and form networks and how the networks generate whole cell functions corresponding to observable phenotypes (Palsson, 2006) • Systems Biomedicine addresses the challenge of translating insights in biological systems to clinical application � 3
Systems Pharmacology • Systems Pharmacology is the application of the concepts of systems biomedicine to pharmacology in order to understand the full effect of a drug • Personalised medicine aims to match each patient with their most beneficial treatment � 4
Disease and Therapy • Identifying key drivers of a disease helps us to design targeted therapies • Drug combinations may help in diseases driven by several altered proteins • In diseases like cancer, we also have to address development of resistance to the applied therapy • Not all the targets are actionable • Not all the therapies need to target the diseased cell: immunotherapy � 5
Importance of signalling networks • Melanoma patients with BRAF mutation show response to BRAF inhibitors Vemurafenib BRAF Proliferation � 6 Note: simplified interaction diagram
Importance of signalling networks • But resistance to treatment eventually develops, leading to relapse Figure 2 in Wagle et al. (2011) shows BRAF-mutant melanoma patient (A) before treatment, (B) after 15 weeks of therapy, and (C) after relapse, after 23 weeks of therapy. Wagle et al., J Clin Inv , 29:22, 2011 � 7
Importance of signalling networks • Colon cancer patients with the same mutation show resistance to treatment because of EGFR feedback loop EGFR Vemurafenib PI3K BRAF AKT Proliferation � 8 Note: simplified interaction diagram
Challenges • Biomedical research faces different challenges: • Noise • Batch effects • Small sample size • Difficult / Expensive experiments • Possible ways of dealing with these: • Well thought and designed experiments • Pool information from different studies • Use of prior knowledge f(x) • Development of mathematical models � 9
Pipeline overview Raw Data f(x) normalize Data Modeling Literature Literature Data Optimizer Bases formalism Omnipath Logic Fitted Question PKN Model Model CellNOpt Cytoscape MaBoSS Structural Experimental Analysis Analysis data compare Visualization Simulation In-silico Network analysis Prediction perturbations � 10 Traynard et al. CPT: Pharmacometrics & systems pharmacology 2017
Biological question • The first step of modelling is to start with a biological question of interest Question • Example: what are the changes in the phosphoproteomic response to the PI3K pathway when a prostate cancer cell goes from being castration sensitive to castration resistant � 11
Network model Data Literature Bases Omnipath Question PKN Cell cycle MYC Caspase9 Caspase8 p38 NFkB OR AND AND AND TNFR p53 OR AND AND Türei et al. Nature Methods 2016 IKK Survival � 12
OmniPath • http://omnipathdb.org graphics by Spencer Phillips Türei, Korcsmáros & Saez-Rodriguez (2016). Nat Methods, 13(12)966-967. � 13
OmniPath • OmniPath is a comprehensive collection of literature curated human signaling pathways • Why Omnipath? number of LMPID 50 interactions PDZBase 100 CA1 ELM 500 Guide2Pharma DOMINO 1000 HPRD-phos 2000 SignaLink3 DeathDomain phosphoELM Signor similarity over dbPTM interactions 0.05 SPIKE 0.10 Macrophage ARN 0.20 PhosphoSite 0.50 http://omnipathdb.org/ DEPOD NRF2ome Türei, Korcsmáros & Saez-Rodriguez (2016) • Available via a webservice or using pypath, a Python module for molecular networks and pathways analysis � 14
OmniPath Türei, Korcsmáros & Saez-Rodriguez (2016) � 15 15
Network model TNFa IGF_1 DHT Stress IL6 EGF IL6R EGFR TNFR IGF1_R RAS Rac JNK Jak PI3K AR MEK p38 Stat3 AKT GSK3 IKK beta ERK1_2 HSP27 catenin mTOR NFkB RPS6 MYC Caspase9 p53 Experimental conditions: stimulated Cell cycle Caspase8 inhibited measured Survival inhibited & measured � 16 Traynard et al. CPT: Pharmacometrics & systems pharmacology 2017
Logic model f(x) Data Modeling Literature Literature Bases formalism Omnipath Logic Question PKN Model � 17
Choice of modelling formalism Physicochemical modeling Causal (logic) modelling The amount of details to include in the model and the mathema&cal formalism used to describe the process should be lead by the biological ques&on (and by available data). Figures from: Saez-Rodriguez J, et al. Annual Rev Biomed Eng , 2015 18
Variety of formalisms • Boolean simulation with synchronous updates • Constrained fuzzy logic • Simulations with multiple time-scales • Logic based ODEs • … � 19
Using logic ODE as modelling formalism Based on ordinary differential equations derived from logic models using a continuous update function (1 − x ) n 1.0 (1 − x ) n + k n f ( x ) = 1 − 1 1+ k n A 0 ≤ k ≤ 1 , n = 3 0.8 dx B = τ B [ f ( x A ) − x B ] dt 0.6 f(x) B 0.4 k=1 0.2 k=0 0.0 0.0 0.2 0.4 0.6 0.8 1.0 x … x i 1 x i 2 x iN dx i x i = τ i ( B i ( f ( x i 1 ) , f ( x i 2 ) , . . . , f ( x iN )) − x i ) ˙ dt x i DM Wittmann, et al., 2009, BMC Systems Biology � 20 F Eduati, 2017, Cancer Research
Using logic ODE as modelling formalism • Easily interpretable parameters strength of regulation j → i is the life-time of species i k ij =0 no edge τ i =0 node not functional >0 stronger interaction >0 higher functionality • Direct derivation from logic rules Generalisation for OR gates B ( f ( x 1 ) , f ( x 2 )) = ... x 1 x 2 0 0 0(1 − f ( x 1 ))(1 − f ( x 2 ))+ 0 1 (1 − f ( x 1 )) f ( x 2 )+ f ( x 1 )(1 − f ( x 2 ))+ 1 0 f ( x 1 ) f ( x 2 ) 1 1 � 21
Collect data Raw Data f(x) normalize Data Modeling Literature Literature Data Optimizer Bases formalism Omnipath Logic Fitted Question PKN Model Model CellNOpt • Objective: obtain data for training logic models • Priority: high number of perturbations � 22
Phospho-proteomics to look at signal transduction Reverse Phase Protein Arrays An&body-based methods: Samples Western Blot low coverage MicroWestern 1000 many condiIons Intracellular Flow Cytometry T argeted MS 100 Protein Microarrays xMAP Mass-spectrometry methods: 10 Label Free MS high coverage Labeled MS few condiIons 1 T arget proteins 1000 1 10 100 Terfve C, Saez-Rodriguez J, Adv. Syst. Biol. , 2012 Saez-Rodriguez J, et al. Annual Rev Biomed Eng , 2015 23
Phospho-proteomics Image showing mass-spectrometry protocol (https:// upload.wikimedia.org/wikipedia/commons/1/1f/ Mass_spectrometry_protocol.png) Credit: By Philippe Hupé [CC-BY-SA-3.0 (http://creativecommons.org/licenses/by-sa/3.0)], via Wikimedia Commons https://pharmchem.ucsf.edu/research/physbio/proteomics � 24
Preparing the data • Normalisation challenges • Boolean logic works with binary values, but measurements are continuous values • CellNOpt ODE works with values between 0 and 1 • Coverage challenges • Not all the nodes in the model may be covered by the measurements • Use of derived measurements • e.g. Kinase activities � 25
Use data for training models Raw normalize Data Experiments Data Optimizer Logic Fitted Question PKN Model Model Omnipath CellNOpt Modeling Data Literature formalism Bases f(x) � 26
www.cellnopt.org � 27
Broad spectrum of modelling formalism with different level of detail + detail , - scope Boolean -- -- quantitative 1 1 steady-state -- -- time series Networks 0 0 t t CellNet time course data? Optimizer n y New sources scarce data? 2 t? >2 t Computable Model large network? specific to data partial e fg ects (cell/time/conditions) rich data? negligible? y ~small network? Terfve C Cokelaer T y n MacNamara A Henriques D y n Gonçalves E Morris MK van Iersel M Lauffenburger DA Saez-Rodriguez J CellNOptR BMC Syst Biol, 6: 133, 2012 CellNOptR CNORfuzzy CNORdt CNORode 2t � 28
PHONEMeS • PHONEMeS (PHOsphorylation NEtworks for Mass Spectrometry) is a method to model signalling networks based on untargeted phosphoproteomics mass spectrometry data and kinase/phosphatase- substrate interactions (Terfve et al. 2015 Nature communications) • We can use it to combine high-throughput data (SWATH phospho-proteomics) with a large scale background network (e.g. Omnipath) � 29
PHONEMeS LNCaP LNCaP-ablated target kinase intermediate kinase measured phosphorite � 30
Fitting example Raw Data f(x) normalize Data Modeling Literature Literature Data Optimizer Bases formalism Omnipath Logic Fitted Question PKN Model Model CellNOpt � 31
Recommend
More recommend
