Welcome to CPS Lunch Forum. It will be a hybrid meeting taking place at Blindern, and it will also be possible to follow the lecture through Zoom. To receive the zoom-link, send an e-mail to ingvihar@uio.no. The meeting is open for everyone!
About
Mai Ha Vu is postdoctoral fellow at ILN, UiO. She is working with the Immunolingo Project, where they try to crack the "grammar" of antigen-antibody binding in an interdisciplinary collaboration between immunology, machine learning, statistics, and linguistics.
Abstract
The idea of using language as a metaphor for biological sequences has a
long history, due to the similar representations of (spoken) natural
language as strings of sounds/letters and biological sequences as
strings of nucleic acids in the case of DNA or a string of amino acids
in the case of proteins. For adaptive immunology, the analogy with
language was most prominently featured in Niels Jerne's 1985 Nobel
lecture.
The adaptive immune system contains billions of highly diverse (~10^14)
and highly specialized adaptive immune receptors that (in a healthy
system) together are capable of recognizing almost any foreign molecule
(antigen) and also do not recognize molecules of the self. There is
then a sense that these adaptive immune receptors are governed by a
'grammar', which determines a given immune receptor's specificity (the
antigens that it can recognize). Similarly, natural language can
generate highly diverse, unbounded number of sentences, governed by a
rules that map between string and meaning. However, the extent to which
the language metaphor can apply to biological sequences and their
usefulness for discovering biological 'grammar' remains an open
question.
Immunolingo is an interdisciplinary project between life sciences,
informatics, statistics and linguistics with the goal of learning these
mapping rules. In this talk, I will talk about the challenges we face
when applying linguistics to biological data and the limits of the
metaphor, and a methodological framework we propose for a more
systematic and hopefully fruitful synthesis of linguistic and machine
learning methods in application to biological data.