Johannes<p><span class="h-card" translate="no"><a href="https://fediscience.org/@FMarquardtGroup" class="u-url mention" rel="nofollow noopener" target="_blank">@<span>FMarquardtGroup</span></a></span> <span class="h-card" translate="no"><a href="https://wisskomm.social/@MPI_ScienceOfLight" class="u-url mention" rel="nofollow noopener" target="_blank">@<span>MPI_ScienceOfLight</span></a></span><br>I recall two presentations that got me hooked on automation in scientific discovery: <br>1) one by Hitachi showcasing the cell culture robots they were developing - massive room-filling devices that could handle and monitor the entire process.<br>2) one by Autodesk on software to design experiments in a way they could be executed by robots. (I also remember that the presenter held rather creepy techno-optimistic views on the genetic editing of humans).</p><p>Cell culture is fun, but there is no denying that a robot could do so much more than my monkey brain and fingers can handle. Yet, while semi-automated microscopy, DNA/RNA analysis prep and lHC are becoming common, all academic labs I’ve been to rely on manual cell culture. I’ve also never encountered full design-excecution-analysis automation as in the Adam/Eve prototype - even though I can see the beauty of it. I wonder why?</p><p><a href="https://fediscience.org/tags/science" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>science</span></a> <a href="https://fediscience.org/tags/AutomatedScientificDiscovery" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>AutomatedScientificDiscovery</span></a></p>
photog.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
A place for your photos and banter. Photog first is our motto
Please refer to the site rules before posting.
Administered by:
Server stats:
244active users
photog.social: About · Profiles directory · Privacy policy
Mastodon: About · Get the app · Keyboard shortcuts · View source code · v4.4.3
#automatedscientificdiscovery
0 posts · 0 participants · 0 posts today
Florian Marquardt<p>A famous example of <a href="https://fediscience.org/tags/AutomatedScientificDiscovery" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>AutomatedScientificDiscovery</span></a> is "Adam the Robot Scientist". </p><p>This is a machine, introduced in 2004 by Ross King and others, that can do biochemistry experiments on its own and smartly pick the next experiment to do. More precisely, it is a room full of robots and automated chemistry, growing yeast cells that have been genetically modified. The goal is to find out which enzymes are important in which parts of the chemistry of yeast. This is a puzzle, since when you switch off a gene and the enzyme it coded for, then you only see indirectly what happens to the yeast (some important substances fail to get produced). It is like a cross-word puzzle, and therefore <a href="https://fediscience.org/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>AI</span></a> can help. Here AI came in the form of the logic programming language <a href="https://fediscience.org/tags/Prolog" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Prolog</span></a>, that can encode all the observations and rules.</p><p>The biggest achievement of the robot scientist is that it can be very clever in selecting the next step, focusing on the most informative experiment. This is essential, since running an experiment is costly and time-consuming.</p><p>The plot I love most in the 2004 paper is the classification accuracy plotted vs the logarithm of the experimentation cost in British pounds. 💷 🤣</p><p>Adam's follow-up robot is called "Eve", doing drug screening.</p><p><a href="https://en.wikipedia.org/wiki/Robot_Scientist" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">en.wikipedia.org/wiki/Robot_Sc</span><span class="invisible">ientist</span></a><br><a href="https://royalsocietypublishing.org/doi/10.1098/rsif.2021.0821" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">royalsocietypublishing.org/doi</span><span class="invisible">/10.1098/rsif.2021.0821</span></a><br>(image below from this article, CC-BY-4.0)</p><p>Read more blog posts in this series: <a href="https://florianmarquardtmastodon.eu.pythonanywhere.com/?account=%40FMarquardtGroup%40fediscience.org&hashtag=AutomatedScientificDiscovery&message=Florian's%20Blog:%20Automated%20Scientific%20Discovery" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">florianmarquardtmastodon.eu.py</span><span class="invisible">thonanywhere.com/?account=%40FMarquardtGroup%40fediscience.org&hashtag=AutomatedScientificDiscovery&message=Florian's%20Blog:%20Automated%20Scientific%20Discovery</span></a></p><p><span class="h-card" translate="no"><a href="https://wisskomm.social/@MPI_ScienceOfLight" class="u-url mention" rel="nofollow noopener" target="_blank">@<span>MPI_ScienceOfLight</span></a></span></p>
Florian Marquardt<p>Automated Scientific Discovery aims to make computers capable of aiding scientific progress even in those aspects that go beyond mere number crunching and conceptually simple data analysis, extending into domains that require more creativity and heuristics. The aims are interpretable solutions, generating new insights, and enabling generalization. </p><p>One pioneering example was an expert system DENDRAL developed for the purpose of interpreting mass spectra. This system was developed in the context of the Viking mars lander mission. A mass spectrum is produced when a molecule is split into fragments and the masses of those fragments are measured. It is a difficult puzzle to try to understand all the possible molecular structures compatible with the given mass spectrum. Feigenbaum, Buchanan, Lederberg, Djerassi and their co-workers figured out very smart algorithms to search through the vast space of possibilities.</p><p><a href="https://en.wikipedia.org/wiki/Dendral" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="">en.wikipedia.org/wiki/Dendral</span><span class="invisible"></span></a></p><p>I teach this example in my lecture, because first it is truly groundbreaking pioneering work, but it also nicely illustrates how a smart representation (here: graphs for molecules) and smart search through combinatorially large spaces are an essential part of AI-aided scientific discovery, alongside domain-specific knowledge and heuristics discovered by the system itself along the way. In the future we will see more of that, combined with the power of neural networks.</p><p>More blog posts: <a href="https://florianmarquardtmastodon.eu.pythonanywhere.com/?account=%40FMarquardtGroup%40fediscience.org&hashtag=AutomatedScientificDiscovery&message=Florian's%20Blog:%20Automated%20Scientific%20Discovery" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">florianmarquardtmastodon.eu.py</span><span class="invisible">thonanywhere.com/?account=%40FMarquardtGroup%40fediscience.org&hashtag=AutomatedScientificDiscovery&message=Florian's%20Blog:%20Automated%20Scientific%20Discovery</span></a></p><p><a href="https://fediscience.org/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>AI</span></a> <a href="https://fediscience.org/tags/AutomatedScientificDiscovery" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>AutomatedScientificDiscovery</span></a></p>
Live feeds
Mastodon is the best way to keep up with what's happening.
Follow anyone across the fediverse and see it all in chronological order. No algorithms, ads, or clickbait in sight.
LoginDrag & drop to upload