<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN"
"http://www.w3.org/TR/REC-html40/loose.dtd">
<html>
<head>
<meta http-equiv="content-type" content="text/html; charset=utf-8">
<title></title>
</head>
<body style="font-family:Arial;font-size:14px">
<p><img alt="" src="cid:e4e8oMBOn_PKM20gSLGTGNS@horde9-pro" style="width: 1000px; height: 150px;"><br>
Dear community,<br>
<br>
Following the tradition, I2PC (Instruct Image Processing Center) is organizing a seminar series on cryoEM and cryoET methods for image processing. This time the seminars and speakers will be announce in our webpage:<br>
<br>
<a href="https://urldefense.com/v3/__https://i2pc.es/i2pc-seminar-series-2024/__;!!Mih3wA!H8pgphlO-_yCcJv42FUWnZG2hfAE-EGZdPfaFBV7mxLKTl97zwv18RGLh9RoOax-kHP_42jjtZpJs2KUMb8VqQ$" target="_blank">https://i2pc.es/i2pc-seminar-series-2024/</a><br>
<br>
Our first seminar will be:<br>
<br>
"<strong>Cryo-forum: A framework for orientation recovery with uncertainty measure with the application in cryo-EM image analysis" presented by Szu-Chi Chung</strong>.<br>
<br>
It will take place on <strong>October 3rd at 10:00 AM Madrid, Paris, Berlin time,</strong> <strong>9:00 AM London time</strong><br>
<br>
The seminar will be online and free to join in the next link:<br>
<br>
<a href="https://urldefense.com/v3/__https://conectaha.csic.es/b/bla-rkh-dqa-rpn__;!!Mih3wA!H8pgphlO-_yCcJv42FUWnZG2hfAE-EGZdPfaFBV7mxLKTl97zwv18RGLh9RoOax-kHP_42jjtZpJs2JdG2G9Eg$" target="_blank">https://conectaha.csic.es/b/bla-rkh-dqa-rpn</a><br>
<br>
You can see an abstract here:<br>
<br>
Cryo-electron microscopy (cryo-EM) shows great potential for determining protein 3D structures. However, the current workflow is slowed down by the complicated process of estimating 3D orientations from 2D projection images, especially as contaminated or low-quality images are challenging to identify in the dataset. Recent deep learning-based approaches aim to accelerate the process by employing amortized inference, eliminating the need for parameter estimation for each image. Despite this, these methods frequently overlook the presence of outliers and may not adequately concentrate on the components used within the network.<br>
<br>
In this presentation, I will introduce a framework designed to recover orientations directly from the acquired 2D projections in an end-to-end manner. Our approach uses a 10-dimensional feature vector to represent an orientation, followed by a Quadratically-Constrained Quadratic Program to obtain the orientation prediction in unit quaternion format with uncertainty statistics. Furthermore, contrastive learning is employed to improve our model's generalization ability by incorporating pairwise distances between images. Finally, we also comprehensively evaluate the design choices in constructing the encoder network, a topic that has not received sufficient attention in the literature. Our numerical analysis demonstrates that our methodology effectively recovers orientations from 2D cryo-EM images in an end-to-end manner. Notably, the inclusion of uncertainty quantification allows for direct clean-up of the dataset at the 3D level.<br>
<br>
<br>
Best regards<br>
<br>
José Luis Vilas<br>
<br type="_moz"></p>
</body>
</html>