Effectiveness of Network Interdiction Strategies to Limit Contagion During a Pandemic

COVID-19 is an infectious disease that has been declared a global public health emergency by the World Health Organization. Besides claiming over 3 million lives worldwide, COVID-19 led to unprecedented disruption in industrial productivity, trading, and global food supply, resulting in loss of livelihood. Despite initial success in curbing the spread of diseases through a lockdown and rapid vaccine development, human lives are threatened by sudden outbreaks from new strains of the virus. This motivates the conceptualization of effective interdiction rules to inform human mobility in a manner that the damage to lives as well as the economy could be minimized. In this work, we present three interdiction rules that employ machine learning-based network inference on daily infected cases to infer the influence of contagion between neighboring zones. The proposed rules leverage network science concepts such as coloring and clustering to attain time-varying partial or complete travel restrictions. Through extensive simulation experiments, we show that these strategies yield lower infection spread than greedy and random migration-based tie elimination approaches as well as a balance between contagion mitigation and economic gain.


I. INTRODUCTION
Coronavirus disease  is an infectious disease that causes mild to severe respiratory illness. It has caused 3.2 million deaths worldwide and led to a prolonged period of uncertainty in public health, livelihood and well-being [1]. In the absence of vaccination and drugs, the nations were exercising social distancing and lockdown to curb the transmission of the virus. The ongoing restrictions has caused a decline in industrial productivity, stock exchange percentage, hike in goods prices and contraction of global GDP. Border and trade restrictions prevent farmers and markets from trading, disrupting global food supply chains [2]. Recent successes in expeditious vaccine developmental efforts have allowed nations to relax these distancing regulations [3].
Despite successes in vaccine development, clinicians, immunologists and epidemiologists have established that the war against COVID-19 is far from over. COVID-19 will continue to exist as a seasonal fixture and continue to claim lives much like the influenza and other coronavirus strains.
The associate editor coordinating the review of this manuscript and approving it for publication was Jad Nasreddine .
Our future depends on how quickly and effectively we immunize ourselves against the evolving strains of the virus [4]. This necessitates continuous vaccine research and rollout endeavors to protect the population from severe symptoms in the event of sudden outbreaks from new strains. It is also imperative to design effective interdiction rule strategies that can leverage statistics of infection to guide human mobility in order to curb contagion with minimum economic ramifications.
spread. Pandey et al. applied the susceptible-exposedinfected-recovered (SEIR) and regression models to study the epidemiological data of COVID-19 in India between a period of January to March, 2020 to infer the reproduction number [6]. Annas et al. incorporated vaccination and social distancing into the calculation of the SEIR parameters for Indonesia [7]. Radulaescu et al. performed a case study on communities in New York to analyse spread dynamics in a age-heterogeneous scenario [8]. Lopez et al. showed through the SEIR epidemic model that social distancing and isolation can cause a manifold decrease in infection spread in the context of Spain and Italy [9]. Tang et al. used SEIR to predict contagion trends in scenarios where the infected individual may cause spread during the incubation period [10].

2) IDENTIFICATION OF FACTORS THROUGH MACHINE LEARNING
Epidemiologists, health workers, clinicians, and social scientists relied on machine learning to identify the different socioeconomic and demographic factors contributing towards pandemic spread [11]. Prediction models are built around the epidemiological and clinical data to pinpoint vulnerable age groups [12]- [14]. Khan et al. applied regression tree analysis, cluster analysis and principal component analysis to study the variations of testing in the identification of confirmed cases [3]. Roy et al. explored the various socioeconomic and demographic factors in the US states affecting the post-lockdown infection spread [15], most-affected job sectors [16], and the social acceptance of mask-wearing, vaccination and distancing in the US [17]. Liu et al. used neural networks to design a nonlinear, self-adaptive dynamic coefficient infectious disease prediction model [18]. Ascani et al. employed auto-regressive model using a weight matrix comprising inverse inter-zone distance to show that smaller economic zones offering scope for interactions through human mobility contribute heavily towards contagion [19] Shinde et al. discuss the advantages and challenges of forecasting COVID-19 spread based on factors such as environmental factors, incubation period, quarantine, age, gender, etc. [20].
Paez et al. performed spatiotemporal analysis on the spatial factors in Spain. They employed spatial Seemingly Unrelated Regressions (SUR) to show the effect of population density, temperature and humidity on the incidence of COVID-19 [21]. Paez also utilize mobility and COVID-19 cases data to establish the high correlation between higher mandatory and discretionary mobility to work and parks and incidence of COVID-19 [22]. Similarly, Noland explored the relationship between mobility and effective reproduction number of the virus [23]. Avineri et al. discussed the challenges arising from the travel restrictions due to the pandemic. They highlighted the importance of the timing of lockdown, by citing global trends showing that early lockdowns allowed nations to continue to 40% mobility while countries that enforced lockdown much later had to enforce tighter lockdowns [24]. Horcher et al. reviewed the literature on information and ticketing technology, covering factors such as queue, space, time, travel permits, etc., to discuss the applicability of social distancing in public transport [25].

3) NEW INTERDICTION AND VACCINE DISTRIBUTION STRATEGIES
Sarlas et al. presented the notion of betweennessaccessibility, a network centrality measure that combines the well-studied concepts of network science and transport geography, namely, betweenness with accessibility. They demonstrated through a case study on Zurich map that the proposed metric identifies critical nodes and links whose interdiction hampers a network's serviceability [26]. Scott et al. presented a metric for identification of critical transport network links, called Network Robustness Index (NRI), measured as the change in travel-time cost associated with rerouting network traffic should that link become unusable. They demonstrate that NRI offers better high planning solutions resulting in improved travel-time savings [27].
The non-pharmaceutical intervention of social distancing has been a primary rule to reduce contagion for a big part of the pandemic. Block et al. [28] and Roy et al. [29] proposed social distancing measures using network science concepts of clustering and homophily to enable individuals in making local decisions to curb spread. With regard to vaccines, there has been scepticism on the part of the public to accept the emerging vaccinations [30]. The national governments are challenged by the task of rapid clinical trials, industrial production of vaccines as well as their distribution at an affordable price [31]. Shim presented a dynamic model to distribute limited vaccines in South Korea [32]. Matrajit et al. presented an age-stratified optimization approach based on deaths, symptomatic infections, and ICU/non-ICU hospitalizations [33]. Roy et al. presented a time-varying vaccination distribution strategy that incorporates socioeconomic and demographic factors to disseminate vaccines among zones [34].

B. CONTRIBUTIONS
It is evident from our discussion in Secs. I-A2 and I-A3 that human interaction via transportation networks contributes greatly to epidemic spread. It is essential to design strategic interdiction measures to limit human interaction by managing mobility across zones. In this work, we study strategies for network interdiction using social network and machine learning techniques. We first employ a well studied random forest-based network inference model, called GENIE3, to derive the strength of influence of contagion between any two zones. We then apply the three proposed interdiction rules, based on network science and optimization principles such as graph coloring, community detection and knapsack problem, to achieve time-varying travel restrictions.
We carry out extensive simulation experiments on the counties of New York City to demonstrate the efficacy of the rules. We show that these rules differ from existing social distancing algorithms because it operates on the daily infected VOLUME 9, 2021 cases of a region and does not require the mobility traces of the individuals (which is not always available). Our analysis shows that the proposed approaches not only outperform greedy and random travel tie elimination methods, but also offers a good mix of contagion mitigation and economic gain. We also analyze the effect of delays in lifting travel restrictions on the overall economy of the affected zones.
This paper is organized as follows. We discuss the preliminaries and proposed approach in Sec. 2. We discuss the data and results of the simulation experiments in Sec. 3. We present the limitations of the study and concluding remarks in Secs. 4 and 5 of the manuscript.

II. APPROACH
Let us review the network science concepts used in this work, followed by the three proposed travel rules.
non-empty set of objects called vertices (or nodes); and E is a (possibly empty) set of 2-subsets of V , called edges [35]. A directed graph has directed edges (u, v) ∈ E, allowing unidirectional information flow from vertex u to v and not necessarily from v to u. In a directed graph, each node u ∈ V has in-neighbors defined as a set of nodes v such that there exists an edge from v into u, i.e., e(v, u) ∈ E. The outneighborhood of u ∈ V consists of nodes v such that there exists an edge u into v, i.e., e(u, v) ∈ E. In a weighted directed graph, (u, v) ∈ E has weight w u,v ∈ [0, 1], measuring the strength of influence of u on v.

Consider a set of geographical sub-regions (termed zones) B.
We define a frequency matrix F ∈ M |B|×|B| (R), such that f i,j ∈ F denotes the number of trips made from zone b i ∈ B to b j ∈ B. We also define a transition matrix A ∈ M |B|×|B| derived from F as follows: each element of the matrix a i,j ∈ A is the probability of making a trip from b i to b j , calculated as . These matrices (F and A) determine the number of people migrating from one region to another based on the inter-zone mobility procedure.

3) EPIDEMIC MODEL
We implement the susceptible-exposed-infected-recovereddeath (SEIRD) model (see Fig. 1) [36], where susceptible (S) class comprises individuals who are not exposed to the infection; upon exposure, they may transfer to the exposed (E) category of asymptomatic or untested individuals, and this transition is controlled by β (that is not a probability). The E class individuals transition to the (tested) infected (I) with probability σ . The individuals in I transition to another state with a probability γ , either recovering (R) or dying (D) with probabilities 1 − α and α, respectively, as shown below.
In the above differential equations, β = p×C, where p and C are the infection probability and contact rates, respectively. Given constants c 1 , c 2 , collision diameter d and population

4) NETWORK INFERENCE
Let us consider a timeline of T days and infection matrix where n is the number of zones and x i ∈ R T is the data corresponding to the u−th zone (i.e., x t u is the measurement of zone u at time t). For any pair of zones u and v, we employ GENIE3 [37] inference model to leverage X in order to gauge how u influences v (u = v), and vice versa. GENIE3 operates under the assumption that any x u is a function f of the expression of x v (u = v) plus some random noise , and can be expressed as: In this equation, X t −u is the matrix containing the t-th measurement of all vectors except x u , i.e., As we illustrate in Fig. 2, GENIE3 applies machine learning on features X t −u to select a subset of features influencing x u . The interaction between zone pairs u and v at any timepoint, w(u, v), is represented as a fully-connected, directed weighted network of V nodes (s.t., |V | = n). In the rest of the paper, we will refer to this directed weighted network as the GENIE3 network or graph.

5) GRAPH COLORING
It is a graph labeling approach in which two nodes sharing a link are assigned different colors [38]. The minimum number of colors needed to color the nodes is called the chromatic number of the graph κ. It is possible to split the graph into κ partitions, where links can only exist between nodes of different partitions, called color classes. We employ the greedy coloring approach [39] where the first node is assigned color 0. The subsequent nodes are assigned an existing color that is not taken up by their neighbors, while a new color is assigned if all the colors are taken by neighbor nodes.

B. PROPOSED APPROACH
We utilize the GENIE3 network inference approach (see Sec. II-A4) on the daily infection numbers of zones in a geographical region to create networks capturing the epidemiological influence of each zone on the other. We present three travel restriction rules that employ network science and optimization principles on the GENIE3 networks to identify  GENIE3 links, which, when removed, can curb contagion. It is noteworthy, the travel restrictions are imposed by eliminating the removable links from the inter-zone migration matrix A that we introduced in Sec. II-A2.

1) COLORING BASED TRAVEL RULE
We propose a simple interdiction rule strategy where at each time t, we calculate the GENIE3 network on the daily infection count of the zones between [t − W , t], G t−W →t . We find the weighted GENIE3 network as follows: Each node u is assigned a weight, z(u), equal to the mean infection numbers over the last d days and replacing the weight of each GENIE3 link (u, v) by w(u, v) × z(u) (v ∈ V ). The intuition behind this step is to attach higher importance to strong GENIE3 influence weight w(u, v), high infection numbers of the source node u, or both.
Given an integral threshold ζ , we temporarily eliminate the directed links e ∈ E(G t−W →t ) if weight w(e) ≤ ζ . These remnant links, called conflict links, represent the strong inter-zone interaction leading to contagion. Consider the simple and weighted GENIE3 graph shown in Figs. 3a and 3b (where the diagonal links, although existent, are not shown in the interest of readability). Fig. 3c shows the conflict network G c for ζ = 7.5. Next, since each link (u, v) in the conflict network represents high likelihood of spread from zone u to v, we create two networks, only preserving the non-conflict links, i.e., (u, v) ∈ E(G\G c ). We apply greedy coloring (see Sec. II-B1) on the (weighted) conflict network and identify the color classes for both networks (κ = 2 color classes as shown in Figs. 4a, 4b). Each link (u, v) in the two networks follow the two ordering rules: (1) u < v and (2) u > v, respectively. Finally, we coalesce the two ordered VOLUME 9, 2021 networks (Figs. 4c) and preserve their links in the migration matrix of the next time window [t, t + W ].

2) KNAPSACK-BASED INTERDICTION RULE
We create a time-varying interdiction rule based on the GENIE3 network G(V , E). Given directed link e = (u, v) ∈ E, we define decision variable X e which denotes the fraction of the time window for which individuals are allowed to move from zone u to zone v. Each link e has an economic gain α(e). min X e∈E We apply linear optimization modeled on fractional knapsack [40], where we drop links e ∈ E with the highest weight w(e), such that economic loss is restricted by a budget B. In Constraint 9, the economic cost associated with a link e = (u, v), is calculated on the basis of GDP of zone u as follows:

3) QUARANTINE ZONES
Let us consider a set of nodes C that experience an outbreak at any given time t with social network G t . We define an outbreak as a sustained rise in daily infection numbers such that x% of the overall population is infected. The most intuitive approach to prevent the spread to non-outbreak zones is to mitigate spread of infection from c ∈ C to the rest of the social network by isolating c; this can be achieved by cutting communication links E from and to c (i.e., E = {(u, v) : (u, v) ∈ E, u = c||v = c}). However, as we discuss in Sec. I, there are economic implications to a lockdown. Moreover, it may be logistically difficult from a policymaking standpoint to isolate the strongly connected links on a short notice. Instead, we propose a community detection approach to identify and isolate the strongly connected communities around each c ∈ C, where the strength of connectivity is measured in terms of the GENIE3 link weights. Unlike the isolation approach, the community based quarantine attempts to eliminate links with low GENIE3 weights (i.e., low interaction).

a: APPROACH
For each c ∈ C, we initialize H c as a graph with a single node c. We iteratively add the node u s.t. (c, u) is the outgoing link with the highest weight. We choose the final configuration H c with the highest cost C, given by: Here, H c is the community of c, while I c is a complement community of c represented by a graph H c with node set given by V (G)\V (H c ). Like all data clustering techniques [41], Eq. 12 maximizes the mean intra-cluster density and minimizes the intra-cluster density between I c and H c . The density of a graph is calculated as: Fig . 5 shows the GENIE3 network where the link thickness is proportional to the GENIE3 weight. Consider an outbreak zone A marked in red. The community approach iteratively includes node D to the outbreak community and removes the links across communities.

b: TIME COMPLEXITY
We employ the greedy strategies to solve the three interdiction rules proposed in this work. Given a graph G(V , E), the greedy coloring based rule (see Sec. II-B1) has a time complexity of O(|V | 2 ) since a node chooses the least color that has not already been used by its neighbor. Greedy fractional knapsack rule (see II-B2) has complexity of O(|E| log |E|) since it ranks the links e ∈ E in the increasing order of value:cost ratio, i.e., w(e) α(e) and selects the links with the least ratio (s.t. X e × α(e) = budget B). Finally, given c ∈ C outbreak zones, the quarantine strategy (see Sec. II-B3) explores |C| neighborhoods incurring a complexity of O(|C| × |V |) = O(|V | 2 ). Note that the proposed approaches are invoked in intervals of several days, allowing the organization deciding the interdiction rule to have enough time and computational resource to execute them.

III. EXPERIMENTAL RESULTS
The results are classified into five subsections: (A) baseline approaches, (B) data collection, (C) Mobility and GDP of New York City, (D) coloring-based interdiction rule, (E) Knapsack and community interdiction rules.
The proposed systems are implemented in Python. We employ the GENIE3 [37] and PuLP [42] libraries for network inference and optimizations operations respectively. We carry out experiments on the 62 counties of New York City, where we consider a high yet heterogeneous initial infected population (ranging between approximately 200 individuals in small counties like Hamilton to 20,000 individuals in large counties like Queens). Unless otherwise stated, we record the mean results for simulation experiments for 50 days over 10 runs. Default simulation parameters are summarized in Table 1.

A. BASELINES APPROACHES
The interdiction rules are implemented by partial and complete removal of migration ties between zone pairs. We compare the proposed interdiction rules against two baselines.

B. DATA COLLECTION
We discuss the NYC map and mobility traces and the NYC (and global) infected and death numbers.

1) MAP GENERATION AND LOCATION IDENTIFICATION
The list of NYC boroughs and districts is acquired from [43], and the latitude and longitude of the 5 boroughs and 62 districts are taken from the Python library for geocoding services, called GeoPy [44]. The distance between any pair of points (i.e., boroughs or districts) on the NYC map is calculated using the geodesic distance function of GeoPy.

2) NYC MOBILITY DATA
We source the mobility data of NYC traffic from NYCOpen-Data [45] -a data repository for fields ranging from city government, education, environment, health to public safety, recreation, social services and transportation. The stated data (spanning a period from 2014 to 2019), collected by the Department of Transportation of New York Metropolitan Transportation Council (NYMTC), has the following fields: ID, road name, source and destination intersecting street name, compass direction, date and time. We use this data to calculate the transition matrix (see Sec. II-A2) that captures the probability of travelling within and across boroughs.

C. MOBILITY AND GDP OF NEW YORK CITY
We discussed in Sec. II-A2 that inter-zone mobility mimics the migration matrix. Although the experiments discussed hereafter are designed around the NYC (districts or) counties, Fig. 6a presents a high-level generalized view of the migration and mobility matrices that are highly similar for the 5 NYC boroughs, (where a borough is a collection of several counties (refer Sec. III-B1 for details). Fig. 6b shows the latitude-longitude coordinates of the NYC counties (shown as circles proportional to their GDP). Fig. 6c shows the 10 counties with highest and least GDP, where the prosperous counties such as, New York, Bronx, Suffolk, Kings, Queens, etc., are fairly clustered together (in Fig. 6b). Note that the migration matrix is not a part of the proposed interdiction strategies and merely employed here to validate them. Therefore, the three approaches can be used on any mobility model.

D. COLORING-BASED INTERDICTION RULE
Before we demonstrate the working of the coloring interdiction rule (introduced in Sec. II-B1), we show the effect of an outbreak in a single zone on the contagion of other zones. Fig. 7a shows the daily infected count when one borough experiences an outbreak (where x = 10% of total population is infected). Note that outbreaks at Manhattan and Brooklyn cause the highest contagion because the two NYC boroughs have the highest population density and population, respectively (see Table 2 for details).
To show the effect of the coloring-based interdiction rule, we first consider different thresholds for link elimination (ζ = 0.4, 0.5, 0.6, 0.7) in the migration matrix. Recall from our discussions in Sec. II-B1, the higher the ζ , the lower will be the degree of travel restriction. Thus, as shown in Fig. 7b, VOLUME 9, 2021  increasing ζ results in higher mean infection numbers on the NYC counties over 10 iterations. We compare the daily infected cases against a scenario where we randomly remove equal number of links in the migration matrix to show that the coloring-based interdiction rule is able to show increasing improvement in curbing contagion over time.

E. KNAPSACK AND COMMUNITY INTERDICTION RULES
Similar to the coloring-based rule, we compare the performance of the knapsack interdiction rule (discussed in Sec. II-B2) against that of random migration based link elimination. Fig. 8 shows that as the permissible economic budget B increases, more migration links are retained resulting in higher total contagion over a period of 50 days. Moreover, the daily infected cases of the proposed knapsack is notably lower than that of random removal.
We discuss in Sec. II-B3 that the community-based rule eliminates links with low inter-zone interaction, offering an improved contagion mitigation versus economic gain tradeoff. Fig. 9a shows that the community approach exhibits lower economic loss due to elimination of low weighing GENIE3 links (as per Eq. 12) across communities, over 50 iterations, while the isolation approach has lower spread. We carry out a simulation, where we create new outbreak zones with x = 10, 20% every 6 days. We compare the daily infection spread for community approach against no travel restriction scenario to show that the proposed approach yields visibly lower contagion (Fig. 9c).
Effect of delay in interdiction rule imposition. The delay due to administrative, political and logistical, etc., reasons also contribute towards contagion. In the community-based interdiction rule, the migration links are eliminated and reinstated by the next day before estimating the new GENIE3 network. In this experiment, we investigate the effect of delay in lifting the travel restrictions (achieved by reinstating the migration links) by a delay of 1, 2, 3, 4 days. Figs. 10a and 10b show that the increase in delay results in   the preservation of more inter-zone migration links, resulting in lower cumulative infection spread.

F. CHOICE OF INTERDICTION STRATEGY
The interdiction strategies applied on the daily infection count informs the level of inter-zone mobility resulting in contagion. This allows the policymakers to modulate the migration matrix in a manner that the interactions resulting in potential contagion can be avoided. To illustrate this point, we show the relationship between the GENIE link and the inter-zone migration weights for the knapsack interdiction strategy for four zone pairs, shown in different panels in Fig. 11. We observe that the migration edge weight (annotated as activity fraction and marked green) is inversely correlated with the GENIE link weights (colored yellow) for all cases.
Comparative study. We compare the performance of coloring-and community-based approaches for varying number of outbreak zones. We consider the following two scenarios: (1) 2 outbreak zones with a high (20%) initial infected fraction and (2) 8 outbreak zones with a moderate (10%) initial infected fraction. For both scenarios, the non-outbreak zones maintain a 2%−5% initial infection. Fig. 12 shows that the community approach that relies on isolation of outbreak zones is more effective when there are fewer (say, 2) outbreak zones experiencing high infection rates. Conversely, the coloring approach is more apt for scenarios where there are several (say, 8) moderate outbreaks. Note that the knapsack interdiction strategy, unlike the other two, allows for fractional inter-zone migration and is particularly useful in scenarios where partial lockdown needs to be enforced.

IV. LIMITATIONS OF THE STUDY
The interdiction rule approaches proposed in this study take the first steps towards curbing contagion, while balancing economic, epidemiological and demographic considerations. There are several limitations in the study. First, in our simulation experiments, we consider outbreak zones to be places with a high daily infection rate. However, incorporating outbreak prediction ahead of time will make the interdiction rules more effective. This motivates the design of machine learning-based preemptive approaches to identify potential outbreak zones. Second, it is imperative to incorporate important factors such as population density and healthcare budget into the interdiction rule approaches. Analogously, we study the effect of delay in lifting travel restrictions i(see Sec. III-E). Our future work will incorporate the effect of delays in the policymaking approaches. Finally, our present approach assumes that the cost of eliminating communication ties between zones is proportional to the inter-zone mobility VOLUME 9, 2021 associated with that tie. It is important to incorporate more elaborate economic metrics into the policymaking.

V. CONCLUSION
We presented three interdiction rule strategies that employ a machine learning-based network inference on daily infected data to mitigate infection spread. While two of the strategies employed complex network concepts such as graph coloring and network clustering, the third is a linear optimization based on the knapsack problem. We carry out simulations on the 62 counties of New York City to demonstrate the efficacy of the proposed schemes. Our analysis shows that the proposed schemes outperform greedy and random baselines for travel tie elimination, and provide a good trade-off between mitigation of contagion and economic losses. This work serves as the basis for elaborate computational models for public interdiction rules that incorporate socioeconomic and demographic factors of geographical regions.
SATYAKI ROY (Member, IEEE) received the Ph.D. degree in computer science from the Missouri University of Science and Technology, Rolla, MO, USA, in 2019. He is currently a Postdoctoral Research Associate with the Department of Genetics, University of North Carolina, Chapel Hill, NC, USA. His research interests include computational biology, network science and optimization, wireless sensor networks, epidemiology, machine learning, and parallel computing.
PREETOM BISWAS is currently pursuing the bachelor's degree with the School of Computing, Informatics, and Decision Systems Engineering, Arizona State University (ASU), Tempe, AZ, USA. He is currently a Math and Programming Enthusiast. His research interests include application of machine learning and graph theoretic approaches to solving real-world problems.
PREETAM GHOSH received the B.S. degree in computer science from Jadavpur University, Kolkata, India, and the M.S. and Ph.D. degrees in computer science and engineering from the University of Texas at Arlington. He is currently a Professor with the Department of Computer Science and directs the Biological Networks Laboratory, Virginia Commonwealth University. He has several federally funded research projects from NSF, NIH, DoD, and US-VHA. His research interests include algorithms, stochastic modeling and simulation, network science, machine learning related approaches in systems biology and computational epidemiology, and mobile computing related issues in pervasive grids that have resulted in more than 170 conference and journal articles. He is currently serves as the Secretary/Treasurer for ACM SIGBio.