Poster abstracts
Poster number 11 submitted by William Ray
Biological data violate the assumptions of Markov models.
William Ray (Nationwide Childrens Hospital and The OSU Biophysics Program)
Abstract:
Markov models rest on a foundational assumption that a system can be modeled as a series of states, where the next state to occur depends only upon the current state (this is called the Markov property). Systems that violate the Markov property cannot be appropriately modeled with Markov approaches.
Biological systems such as proteins, violate the foundational Markov property. The fitness contribution of any given residue in a protein does not depend solely on the preceding residue in the sequence. Yet Biology has a long-standing tradition of trying to apply Markov Models to determining fitness in proteins.
This is much like looking for one's car keys under the lamp near the store-front, because it's convenient to look there, rather than searching out in the dark parking lot where one dropped them.
We propose that a different, more sophisticated variety of Graphical Probabilistic Model called a Conditional Random Field is more appropriate to modeling protein fitness from sequence.
References:
William C. Ray, “The Bio/Life-Sciences need better visualization of statistical network structures”, Dagstuhl Reports vol 8.4, pp 61. Edited by Jan Aerts, Nils Gehlenborg, Georgeta Elisabeta Marai and Kay Katja Nieselt. 2018.
Keywords: Proteins, Modeling, Statistics