<html>
<head>
<title></title>
</head>
<body style="font-family:Arial;font-size:14px">
<p>Hola a todos,<br>
<br>
El Martes 27 de Marzo se viene la cuarta -y ultima por ahora- de las*<br>
"Discusiones informales entre Biologos y Matematicos"*<br></p>
<div>a las 14:30, en el aula 2 del Pab. I (FCEyN-UBA)</div>
<p>con charlas de* *Morten Nielsen e Ignacio Sanchez.<br></p>
<div><a href="http://cms.dm.uba.ar/actividades/biomat" target="_blank">http://cms.dm.uba.ar/<wbr>actividades/biomat</a><br>
<br>
(Recuerdo que esta actividad esta relacionada con la primer Conferencia<br>
Internacional de Investigacion Interdisciplinaria<br>
<a href="http://cms.dm.uba.ar/actividades/ci3/" target="_blank">http://cms.dm.uba.ar/<wbr><wbr>actividades/ci3/</a><br>
de la cual nuestra facultad sera anfitriona, en dos semanas:<br>
en particular el dia lunes 9 de Abril a las 14 horas habra una jornada<br>
especial con apertura por Lino Bara~nao, charlas de Alberto Kornblihtt,<br>
Gabriel Rabinovich y el medalla Fields Steve Smale,e introduccion de los<br>
oradores por Jorge Aliaga.)</div>
<p><wbr><wbr><b> <br>
Morten Nielsen:</b> Introduction to Blosum matrices and MHC peptide binding predictions.<br>
In my talk, I will give an introduction to two central topics of<br>
bioinformatics prediction method development. First, I will introduce the<br>
Blosum matrices, and next describe some of the critical pitfalls of<br>
developing and estimating the performance accuracy of bioinformatics<br>
prediction methods.Blosum matrices form an integrated part of most<br>
bioinformatics applications including sequence alignment, homology<br>
modeling, local structure prediction and receptor-ligands binding<br>
prediction. Here, I will describe, using a simple toy-model system, how<br>
these matrices are derived, and illustrate how they can be applied to infer<br>
biological information in sequence data. At the heart of bioinformatics<br>
lies the development of methods for prediction of the behavior of<br>
biological systems. One such example is the prediction of peptide binding to the MHC molecule. I will give an overview of some of the methods we have developed in our research to characterize this binding event, and illustrate how serious caution must be taken in general when developing predictions methods and in particular when assessing their predictive performance due to potential redundancy in the data use to train and evaluating the methods. I will demonstrate how one can significantly overestimate performance by not dealing with this issue in a proper manner, and describe how, according to me, proper method development and evaluation should be performed. Finally, I will relate these observations to the work on MHC peptide binding described by Quan-Wu Xiao.<br>
<br>
<font><b>Ignacio Sanchez</b></font><font>:</font> Could you be more specific, please?<br>
Most cell functions depend on the establishment of a certain set of protein-protein interactions among the mony possible sets. We are tackling this specificity problem with the help of information theory. </p>
</body>
</html>