High dimensional random walks can appear low dimensional: Application to influenza H3N2 evolution

Area of Science:

• Immunology
• Virology
• Computational Biology

Background:

• Antibodies provide cross-immunity against infections through specific antigen binding.
• Antigenic space models this cross-immunity, with past studies suggesting low dimensionality.
• Influenza evolution is a key area of immunological research.

Purpose of the Study:

• To re-evaluate the dimensionality of antigenic space using influenza hemagglutination data.
• To propose a new model for influenza evolution as a Gaussian random walk.
• To reconcile discrepancies between low-dimensional and high-dimensional models of antigenic space.

Main Methods:

• Modeling influenza evolution as a Gaussian random walk.
• Analyzing hemagglutination data to infer dimensionality.
• Comparing results with principal component analysis (PCA) and multidimensional scaling (MDS).

Main Results:

• Hemagglutination data is consistent with a high-dimensional antigenic space.
• Gaussian random walks can appear low-dimensional under standard analyses like PCA and MDS.
• Prior studies likely underestimated dimensionality due to analytical methods.

Conclusions:

• Antigenic space for influenza is likely high-dimensional.
• High dimensionality implies a reduced role for pre-existing host immunity in influenza dynamics.
• This has significant implications for epidemiological modeling and vaccine strategies.