Social network and disease spread Laurens Bakker, Philippe Giabbanelli Outline ▪ What is a social network? ▪ Measures ▪ Disease spread ▪ Three case studies L Bakker, P Giabbanelli Social networks and disease spread 1 What is a social network? How does it form? L Bakker, P Giabbanelli Social networks and disease spread 2 L Bakker, P Giabbanelli Social networks and disease spread 3 But really, how does it form? People go places… L Bakker, P Giabbanelli and meet in the process Social networks and disease spread 4 But really, how does it form? People want things… L Bakker, P Giabbanelli and use others Social networks and disease spread 5 But really, how does it form? People have things in common… and express their commonalities L Bakker, P Giabbanelli Social networks and disease spread 6 What is a social network? How does it form? L Bakker, P Giabbanelli Social networks and disease spread 7 Fluffy theories L Bakker, P Giabbanelli Social networks and disease spread 8 If we want to do science… L Bakker, P Giabbanelli Social networks and disease spread 9 we need something with teeth! L Bakker, P Giabbanelli Social networks and disease spread 10 Network definition • Actor => Vertex/Node – Boundary • Connection => Edge/Link – Interaction – Dynamic social networks • Observability – Degree (Dombrowski 2007) L Bakker, P Giabbanelli Social networks and disease spread 11 Measures L Bakker, P Giabbanelli Social networks and disease spread 12 Motifs – Clustering – Average distance – Degree distribution Global Degree distribution Average distance Clustering Motifs Local (Giabbanellli 2011) L Bakker, P Giabbanelli Social networks and disease spread 13 Motifs – Clustering – Average distance – Degree distribution 2 1 1 0 3 0 2 Given a graph G… a motif is a subgraph that appears at a ‘very’ different frequence in G than in S. 0 and a set S of random graphs of the same size and average degree, (Milo 2004) L Bakker, P Giabbanelli Social networks and disease spread 14 Motifs – Clustering – Average distance – Degree distribution L Bakker, P Giabbanelli Social networks and disease spread 15 Motifs – Clustering – Average distance – Degree distribution For a given node i , we denote its neighborhood by Ni. The clustering coefficient Ci of i is the edge density of its neighborhood. Here, there are two edges between nodes in Ni. Ci = 2.2/(5.4) = 0.2 If a graph has ithigh clustering coefficient, then there arei-1) communities At most, would be a complete graph with Ni.(N edges. (i.e., cliques) in this graph. People tend to form communities so they are common in social networks. L Bakker, P Giabbanelli Social networks and disease spread 16 Motifs – Clustering – Average distance – Degree distribution The distance is the number of edges to go from one node to another. The average distance is the average of the distance between all pairs of nodes. L Bakker, P Giabbanelli Social networks and disease spread 17 Motifs – Clustering – Average distance – Degree distribution The average distance l is: ∙ small if l∝ln(n) ∙ ultrasmall if l∝ln(ln(n)) (Newman 2003) (Cohen 2003) L Bakker, P Giabbanelli Social networks and disease spread 18 Motifs – Clustering – Average distance – Degree distribution History (the Hype) • Milgram (Milgram 1969) – Small world • Watts & Strogatz (Watts 1998) – “Small Worlds” & “6 Degrees” • Barabasi & Albert (Barabasi 1999) – Power Law (scale free) • Newman (Newman 2003) – Review L Bakker, P Giabbanelli Social networks and disease spread 19 Motifs – Clustering – Average distance – Degree distribution Many measured phenomena are centered around a particular value. (Newman 2005) L Bakker, P Giabbanelli Social networks and disease spread 20 Motifs – Clustering – Average distance – Degree distribution Many measured phenomena are centered around a particular value. There also exists numerous phenomena with a heavy-tailed distribution. lets plot it on a log-log scale (Newman 2005) L Bakker, P Giabbanelli Social networks and disease spread 21 Motifs – Clustering – Average distance – Degree distribution We also say that thisnumerous distribution follows awith power-law, with exponent α. There exists phenomena a heavy-tailed distribution. The equation of a line is p(x) = -αx + c. Here we have a line on a log-log scale: ln p(x) = -α ln x + c apply exponent e c -α p(x) = ecx (Newman 2005) L Bakker, P Giabbanelli Social networks and disease spread 22 Motifs – Clustering – Average distance – Degree distribution We say that this distribution follows a power-law, with exponent α. computer files people’s incomes Keep in mind that this is quite common. moon craters visits on web pages (Li 2005) L Bakker, P Giabbanelli Social networks and disease spread 23 Disease spread L Bakker, P Giabbanelli Social networks and disease spread 24 Thresholds – Variations – Immunization A ‘threshold’ is the extent to which a disease must be infectious before you can’t stop it from spreading in the population. Very famous claim: scale-free networks have no thresholds! It will spread! (Wikipédia: modèles compartimentaux en épidémiologie) L Bakker, P Giabbanelli Social networks and disease spread 25 Thresholds – Variations – Immunization Very famous claim: scale-free networks have no thresholds! It will spread! « in a scale-free network there is no epidemic threshold thus eliminating a sexually transmissible disease is impossible » (Kretschmar 2007, opening of Networks in Epidemiology) That’s actually sort of false… …it needs additional conditions, that may not exist. L Bakker, P Giabbanelli Social networks and disease spread 26 Thresholds – Variations – Immunization Depending on the diseases, there are several epidemiological classes: infected (I), recovered (R), carriers (C)… It may be interesting to see how the properties of the network influences the number of individuals in each class over time. order randomness (Kuperman 2001; Crepey 2006) L Bakker, P Giabbanelli Social networks and disease spread 27 Thresholds – Variations – Immunization There are four broad approaches (Giabbanelli 2011). Is the disease spreading at the same time? Yes We can immunize anybody We must follow social links Global competitive Global preventive = network game = separator problem NP-hard NP-complete (Kostka 2008) (Rosenberg 2001) Local competitive Local preventive Agents that fight… …and explore (Giabbanelli 2009) L Bakker, P Giabbanelli No (Stauffer 2006) Social networks and disease spread 28 Case Study #1 Measuring what matters L Bakker, P Giabbanelli Social networks and disease spread 29 Example #1: Social networks Measure: distance Property: average distance L Bakker, P Giabbanelli Social networks and disease spread 30 Example #2: Obesity map Measure: Centrality L Bakker, P Giabbanelli Social networks and disease spread 31 Example #3: Backbone network We do not care about clustering or whether the network is scale-free. (Giabbanelli 2010) Measure betweenness and average distance. L Bakker, P Giabbanelli Social networks and disease spread 32 What can we measure in a network? Network Process Measures Social network Disease spread Average distance Factors incluencing obesity Obesity level Centrality Backbone network L Bakker, P Giabbanelli Deploying equipment Betweenness centrality Average distance Social networks and disease spread 33 How do we find out what we should measure? ▪ Know the properties of the network you are studying. → Network analysis ▪ Generate many of them using appropriate stochastic models. → Network generation ▪ Record several measures, and the value of the outcome process. → Possibly optimization ▪ Analyze which measures are linked to the outcome. → Data mining L Bakker, P Giabbanelli Social networks and disease spread 34 Case Study #2 Health & Social Networks L Bakker, P Giabbanelli Social networks and disease spread 35 « People are interconnected, and so their health is interconnected. » « … there has been growing conceptual and empirical attention over the past decade to the impact of social networks on health. » (Smith 2008) Christakis&Fowler have used social networks to show that people are correlated in weight status, smoking, and… happiness! http://www.ted.com/talks/lang/eng/nicholas_christakis_the_hidden_influence_of_social_networks.html L Bakker, P Giabbanelli Social networks and disease spread 36 The basic idea A long imbalance between energy intake&output yields obesity. What spread between people are behaviours impacting intake&output. Eating L Bakker, P Giabbanelli Exercising Social networks and disease spread 37 How we modelled it We used social networks. Each individual has a level of physical activity and an energy intake. L Bakker, P Giabbanelli Social networks and disease spread 38 How we modelled it We also modelled human metabolism. L Bakker, P Giabbanelli Social networks and disease spread 39 Results from Phase 1 L Bakker, P Giabbanelli Social networks and disease spread 40 Results from Phase 1 Presented at ICO 8.6% acceptance Positive reactions Journal on its way L Bakker, P Giabbanelli Social networks and disease spread 41 Case Study #3 Homeless in the tri-cities L Bakker, P Giabbanelli Social networks and disease spread 42 Homeless in the Tri-Cities (I) • Hope for Freedom Society • Vertex definition – Boundary: existence of client file • Edge definition – Interaction: co-observation • Time! – Connection: repeated interaction L Bakker, P Giabbanelli Social networks and disease spread 43 Homeless in the Tri-Cities (II) • Descriptives: – 2 years – ~250 actors – ~3000 observations • Statistical Models – Static: PNET = ERGM = logit p* (Hunter 2006) – Dynamic: SIENA (Snijders 2006) L Bakker, P Giabbanelli Social networks and disease spread 44 References Barabasi 1999 AL Barabasi, R Albert, Emergence of Scaling in Random Networks, Science, 1999 Cohen 2003 R Cohen, S Havlin, Scale-free networks are ultrasmall, Physical Review Letters, 2003. Crepey 2006 P Crepey et al, Epidemic variability in complex networks, Phys. Rev. E, 2006. Drombrowski 2007 K Dombrowski, R Curtis, SR Friedman, Injecting drug user network topologies and infectious disease tranmission: suggestive findings, Working Paper 2007 L Bakker, P Giabbanelli Social networks and disease spread 45 References Giabbanelli 2009 PJ Giabbanelli, Self-improving immunization policies for complex networks, MSc Thesis@SFU, 2009 Giabbanelli 2010 PJ Giabbanelli, Impact of complex network properties on routing in backbone networks, CCNet 2010 (IEEE Globecom) Giabbanelli 2011 PJ Giabbanelli, JG Peter, Complex networks and epidemics, TSI, 2011, to appear. Hunter 2006 D Hunter, Exponential Random Graph Models for Network Data, Talk, 2006, http://www.stat.psu.edu/~dhunter/talks/ergm.pdf L Bakker, P Giabbanelli Social networks and disease spread 46 References Kostka 2008 J Kostka et al., Word of Mouth : Rumor Dissemination in Social Networks, Lecture Notes in Computer Science, 2008. Kretzschmar 2007 M Kretzschmar, J Wallinga, Networks in Epidemiology, Mathematical Population Studies, 2007 Kuperman 2001 M Kuperman, G Abramson, Small World Effect in an Epidemiological Model, Physical Review Letters, 2001. Li 2005 L Li et al., Towards a Theory of Scale-Free Graphs : Definition, Properties and Implications, Internet Mathematics, 2005. L Bakker, P Giabbanelli Social networks and disease spread 47 References Milgram 1969 J Travers, S Milgram, An Experimental Study of the Small World Problem, Sociometry, 1969 Milo 2004 R Milo, et al., Superfamilies of Evolved and Designed Networks, Science, 2004. Newman 2003 MEJ Newman, The structure and function of complex networks, SIAM Review, 2003. Newman 2005 MEJ Newman, Power laws, Pareto distributions and Zipf’s law, Contemporary Physics, 2005. Rosenberg 2001 AL Rosenberg, Graph Separators, with Applications, Kluwer Academic, 2001 L Bakker, P Giabbanelli Social networks and disease spread 48 References Smith 2008 KP Smith, NA Christakis, Social networks and health, Annu Rev Social, 2008 Snijders 2006 TAB Snijders, Statistical Methods for Network Dynamics, Proceedings of the XLIII Scientific Meeting of the Italian Statistical Society, 2006 Stauffer 2006 AO Stauffer et al, A dissemination strategy for immunizing scale-free networks, Phys. Rev. E, 2006. Watts 1998 DJ Watts, SH Strogatz, Collective dynamics of 'small-world' networks, Nature, 1998 L Bakker, P Giabbanelli Social networks and disease spread 49