Ishanu Chattopadhyay PRO
ML | Data Science Biomedical Informatics | Social Science | Assistant Professor
Large Science Models for Predictive Biosurveillance of Zoonotic Emergence
Ishanu Chattopadhyay, PhD
Asst Professor, Biomedical informatics & Computer Science
University of Kentucky
ishanu_ch@uky.edu
CURE-KY: Infectious Research Day 2026
March 19 2026
Are emergence events predictable?
- 1918 onward: classical swine H1N1 persists in pigs [1,2]
- 1979: Eurasian avian-like swine H1N1 becomes established in European swine [2]
- 1998: North American triple-reassortant swine viruses emerge [3]
- 2008-2009: pandemic genome assembled in swine; 6 segments from North American triple-reassortant swine viruses, 2 segments (NA, M) from Eurasian avian-like swine [3]
- Mar-Apr 2009: human outbreak recognized in Mexico and the U.S. [3,4]
- Jun 11, 2009: WHO declares pandemic [5]
- Memoli MJ, Tumpey TM, Jagger BW, Dugan VG, Sheng ZM, Qi L, Kash JC, Taubenberger JK. An early “classical” swine H1N1 influenza virus shows similar pathogenicity to the 1918 pandemic virus in ferrets and mice. Virology. 2009;393(2):338-345. doi:10.1016/j.virol.2009.08.021.
- Dunham EJ, Dugan VG, Kaser EK, Perkins SE, Brown IH, Holmes EC, Taubenberger JK. Different evolutionary trajectories of European avian-like and classical swine H1N1 influenza A viruses. Journal of Virology. 2009;83(11):5485-5494. doi:10.1128/JVI.02565-08.
- Smith GJD, Vijaykrishna D, Bahl J, Lycett SJ, Worobey M, Pybus OG, Ma SK, Cheung CL, Raghwani J, Bhatt S, Peiris JSM, Guan Y, Rambaut A. Origins and evolutionary genomics of the 2009 swine-origin H1N1 influenza A epidemic. Nature. 2009;459(7250):1122-1125. doi:10.1038/nature08182.
- Centers for Disease Control and Prevention. Outbreak of swine-origin influenza A (H1N1) virus infection, Mexico, March-April 2009. MMWR Morb Mortal Wkly Rep. 2009;58(17):467-470.
- World Health Organization. DG Statement following the meeting of the Emergency Committee. 11 June 2009.
the 2009 swine-flu pandemic
1918
1979
1998
2008
2008
2009
Mar
Jun
Lets train a machine learning model to recognize human vs swine Influenza HA proteins
High confidence generally
\textrm{Emergence Risk} = \frac{1}{\textrm{classification accuracy}}
\textrm{Emergence Risk} = \frac{1}{\textrm{classification accuracy}}
Data source: NCBI and GISAID HA sequences collected between 1999 and 2016 globally (HA protein fasta)
Spatial Risk over time
The Signal is clear!
However, Not very actionable
Data source: NCBI and GISAID HA sequences collected between 1999 and 2016 globally (HA protein fasta)
Pr(x \rightarrow y)
Calculate the Quantitative Odds of a strain \(x\) giving rise to strain \(y\) in the wild
Preempt which specific strain is poised to cross the edge of emergence
Hemaglutinnin (HA)
Neuraminidase
Mediates Cellular Entry
Surface structures maximally involved in host interaction
Mediates Cellular Exit
Large Science Models
Learn the emergent constriants from data
Example:
Adjacent sequence blocks do not map to adjacent spatial regions after folding.
-
Non-collinearity: residues that are far apart in sequence can be:
-
physically adjacent in 3D
-
functionally coupled (e.g., receptor binding)
-
-
Functional dependence is long-range:
-
Mutations in HA1 (head) can affect HA2 (fusion machinery)
-
Antigenicity depend on global structural constraints, not local sequence neighborhoods
-
Functional coupling
Statistical association in random perturbations
Building A Large Science Model
NCBI + GISAID
>200,000 HA/NA sequences
Large Science Models: Mathematical Framework
\begin{aligned}
\text{Observables:} \quad & \color{yellow}X = \{x^1, \ldots, x^N\}, \overbrace{x^i \in \Sigma^i}^{\text{finite alphabet}} \\
{\color{gray}\text{Notation:} }\quad & \color{gray} x^{-i} = \{x^j : j \ne i\}\\
\text{Crosstalk:} \quad & \forall i \ P(x^i) = \color{red} f_i(x^{-i}) \\
\text{System state:} \quad & \color{Cyan} \psi = \bigotimes_{i=1}^N \psi^i, \quad \psi^i \in \mathscr{D}(\Sigma^i) \cup \varnothing \\
%\textbf{Degenerate case:} \quad & \psi^i \text{ is a delta distribution (fully observed)} \\
{\color{gray}\text{Notation:} }\quad & \color{gray}\psi^{-i} = \bigotimes_{j \ne i} \psi^j
\end{aligned}
| 222 | 223 | 224 | --- | 560 | |
|---|---|---|---|---|---|
| strain 1 | |||||
| strain 2 | |||||
| --- | |||||
| strain m |
observables
samples
Distributions over alphabet \(\Sigma^i\)
\phi = \bigotimes_{i=1}^N \phi^i, \quad \phi^i(\psi^{-i}) \in \mathscr{D}(\Sigma^i) \\
Individual Predictor (CIT)
cross-talk
\phi(\psi) \vert \vert \psi
Tension between predicted and observed distribution drives change
Example: HA Site 223 on Influenza A
\(\psi^i\)
| K | G | Y | S | T |
\Sigma^i
Digital Twin
\phi^i(\psi^{-i}) \sim \widetilde{\psi}^i
\(\phi\) estimates \(\psi\)
population
individual
estimate is always a non-empty non-degenerate distribution
missing observation
Large Science Models: Properties
\left \lvert \ln \frac{\Pr(\psi\to \psi')}{\Pr(\psi' \rightarrow \psi')}\right \rvert \le \beta\,\theta(\psi,\psi')
Large Deviation Bound*
This bound connects ``closeness'' of strains to the odds of perturbing from one to the other, bridging geometry to dynamics
(Sanov's Theorem, Pinkser's Inequality)
persistence probability
const. scaling as \(N^2\)
Pr(\psi \rightarrow \psi')
\(\psi\)
\(\psi'\)
\(\theta\)
LSM-Distance Metric*
\theta(\psi,\psi') \triangleq \frac{1}{N}\sum_{i=1}^{N} \sqrt{D_{JS}\Bigl(\phi^i(\psi^{-i}) \vert \vert \phi^i(\psi'^{-i})\Bigr)}
where \(D_{JS}(P\vert \vert Q)\) is the Jensen-Shannon divergence.
Distance/similarity depends on the strains AND the background circulation.
Smaller distance \( \Rightarrow \) Higher odds of one jumping to the other
H1N1 2023 Influenza A HA
- Set of conditional inference trees (CIT)
- Strict statistical guarantees: quantifies inference uncertainty
- Each tree models exactly one variable as a function of potentially all other variables
- Non-leaf nodes are "hyperlinked" to other trees
Recursive LSM forest: hyperlinked nodes capturing emergent macro-structures
Large Science Models
Learn the emergent constriants from data
H3N2 2021 Influenza A HA
- Set of conditional inference trees (CIT)
- Strict statistical guarantees: quantifies inference uncertainty
- Each tree models exactly one variable as a function of potentially all other variables
- Non-leaf nodes are "hyperlinked" to other trees
Recursive LSM forest: hyperlinked nodes capturing emergent macro-structures
Large Science Models
Learn the emergent constriants from data
LSM distance \(\neq \) Edit Distance
Same sequence/strain pair, but dfferent background (times of collection)
LSM distance
Small edit distance (5-8) and large LSM distance
Seasonal predictions for Influenza A.
The Vaccine Strain Selection Problem for the seasonal epidemic
Flu vaccine strain recommendation timeline
-
Year-round: WHO GISRS laboratories [1] collect and characterize circulating viruses worldwide.
-
February: WHO issues recommendation for next Northern Hemisphere season.
-
September: WHO issues the recommendation for the next Southern Hemisphere season.
-
March (U.S.): FDA/VRBPAC reviews global and U.S. data and issue recommendation
-
~6–9 months before distribution: manufacturers produce vaccine after strain selection.
Feb
Sep (Southern H.)
- This is essentially a prediction problem
[1] World Health Organization. GISRS. https://www.who.int/initiatives/global-influenza-surveillance-and-response-system
Can we improve?
x_\star^{t+\delta} = \argmin_{y \in \cup_{\tau \leqq t} H^\tau} \left ( \sum_{x\in H^t} \theta_{m_t}(x,y) - \vert H^t \vert A \ln Pr(y \rightarrow y) \right )
Likely strain at time \(t+\delta\)
Set of strains circulating strains at time \(\tau\)
LSM distance of \(x\) to \(y\)
\(A\): constant depending on length of genome
Seasonal predictions for Influenza A.
The Vaccine Strain Selection Problem for the seasonal epidemic
Solve the mathematical optimization problem
"Compute the most likely strain to arise in time \(t+\delta\) given current circulation"
Seasonal predictions for Influenza A.
The Vaccine Strain Selection Problem for the seasonal epidemic
LSM does better with edit-distance mismatch from realized circulation
Caveat: Edit distance closeness might not always reflect immunological response
Measure of Emergence Potential
\rho_t(x) \triangleq -\log \min_{\begin{subarray}{c}y,z \in H^t \\ r \in L_\mathcal{H},\\s \in L_\mathcal{N}\end{subarray}} \sqrt{\theta_r\left (x^{\mathcal{H}},y^{ \mathcal{H} }\right ) \theta_{s}\left (x^{\mathcal{N}},z^{\mathcal{N}}\right )},
Neuraminidase contribution
Hemagglutinin contribution
current animal circulation
Protein-specific Lsms
Which animal strain has the highest odds of giving rise to a strain similar to the circulating humans population?
- Same subtype doesnt need to pre-exist in humans
- Very few observations suffice in the animal circulation
- Fewer strains increase quantifiable uncertainty
Optimization Problem
slow (months), quasi-subjective, expensive
*https://www.cdc.gov/flu/pandemic-resources/monitoring/irat-virus-summaries.htm
24 strains assessed in 14 years
~10,000 strains collected annually
CDC: Influenza Risk Assessment Tool* (IRAT) scoring for animal strains
10 dimensions of assessment
IRAT replication by LSM
(automated, in seconds)
\textrm{Regression Eqn.} y = 0.47x + 2.77\\
R = 0.72\\
\textrm{p-value} = 0.0001
Stamping Out the Next Pandemic **Before** The First Human Infection
BioNorad
\frac{\delta \theta(x,y)}{\delta y}
Which sites are driving prediction?
Estimate how LSM distances perturb as we perturb individual sites
Re-drawing the Phylogeny
LSM-distance induces a new phylogeny
phylogeny with post-2020 sequences
Take-home Message
Emergence is not random—it is the delayed observation of structured, constrained evolution that can be inferred from data.
-
PREEMPT: PREventing EMerging Pathogenic Threats
-
MAGICS: Methodological Advancements for* Generalizable Insights into Complex Systems*
*The views, opinions, and/or findings expressed are those of the author(s) and should not be interpreted as representing the official views or policies of DARPA or the U.S. Government.
Acknowledgement
Prof. Li Feng
William Robert Mills Chair in Equine Infectious Disease, MIMG
Prof. Saurabh Chattopadhyay
Microbiology, Immunology, and Molecular Genetics
Collaborators
cureKY_research_day
By Ishanu Chattopadhyay
