Google DeepMind and their new startup Isomorphic Labs, are expanding into drug discovery. They developed AlphaFold3 as their model to accelerate drug discovery and create demand from big pharma. They already signed Novartis and Eli Lilly for $3 billion - Google\u2019s becoming a pharma company! (https://www.isomorphiclabs.com/articles/isomorphic-labs-kick...)
AlphaFold3 is a biomolecular structure prediction model that can do three main things: (1) Predict the structure of proteins; (2) Predict the structure of drug-protein interactions; (3) Predict nucleic acid - protein complex structure.
AlphaFold3 is incredibly important for science because it vastly accelerates the mapping of protein structures. It takes one PhD student their entire PhD to do one structure. With AlphaFold3, you get a prediction in minutes on par with experimental accuracy.
There\u2019s just one problem: when DeepMind published AlphaFold3 in May (https://www.nature.com/articles/s41586-024-07487-w), there was no code. This \nbrought up questions about reproducibility (https://www.nature.com/articles/d41586-024-01463-0) as well as complaints from the scientific community (https://undark.org/2024/06/06/opinion-alphafold-3-open-sourc...).
AlphaFold3 is a fundamental advance in structure modeling technology that the entire biotech industry deserves to be able to reap the benefits from. Its applications are vast, including:
- CRISPR gene editing technologies, where scientists can see exactly how the DNA interacts with the scissor Cas protein;
- Cancer research - predicting how a potential drug binds to the cancer target. One of the highlights in DeepMind\u2019s paper is the prediction of a clinical KRAS inhibitor in complex with its target.
- Antibody / nanobody to target predictions. AlphaFold3 improves accuracy on this class of molecules 2 fold compared to the next best tool.
Unfortunately, no companies can use it since it is under a non-commercial license!
Today we are releasing the full model trained on single chain proteins (capability 1 above), with the other two capabilities to be trained and released soon. We also include the training code. Weights will be released once training and benchmarking is complete. We wanted this to be truly open source so we used the Apache 2.0 license.
Deepmind published the full structure of the model, along with each components\u2019 pseudocode in their paper. We translated this fully into PyTorch, which required more reverse engineering than we thought!
When building the initial version, we discovered multiple issues in DeepMind\u2019s paper that would interfere with the training - we think the deep learning community might find these especially interesting. (Diffusion folks, we would love feedback on this!) These include:
- MSE loss scaling differs from Karras et al. (2022). The weighting provided in the paper does not downweigh the loss at high noise levels.
- Omission of residual layers in the paper - we add these back and see benefits in gradient flow and convergence. Anyone have any idea why Deepmind may have omitted the residual connections in the DiT blocks?
- The MSA module, in its current form, has dead layers. The last pair weighted averaging and transition layers cannot contribute to the pair representation, hence no grads. We swap the order to the one in the ExtraMsaStack in AlphaFold2. An alternative solution would be to use weight sharing, but whether this is done is ambiguous in the paper.
More about those issues here: https://github.com/Ligo-Biosciences/AlphaFold3
How this came about: we are building Ligo (YC S24), where we are using ideas from AlphaFold3 for enzyme design. We thought open sourcing it was a nice side quest to benefit the community.
For those on Twitter, there was a good thread a few days ago that has more information: \nhttps://twitter.com/ArdaGoreci/status/1830744265007480934.
A few shoutouts: \nA huge thanks to OpenFold for pioneering the previous open source implementation of AlphaFold\nWe did a lot of our early prototyping with proteinFlow developed by Lisa at AdaptyvBio we also look forward to partnering with them to bring you the next versions!\nWe are also partnering with Basecamp Research to supply this model with the best sequence data known to science. \nMatthew Clark (https://batisio.co.uk) for his amazing animations!
We\u2019re around to answer questions and look forward to hearing from you!"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: An open-source implementation of AlphaFold3"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/Ligo-Biosciences/AlphaFold3"}},"_tags":["story","author_EdHarris","story_41448439","show_hn"],"author":"EdHarris","children":[41448800,41448948,41448989,41449014,41449322,41449933,41450513,41450570,41451243,41453870,41454999,41457339],"created_at":"2024-09-04T17:44:17Z","created_at_i":1725471857,"num_comments":37,"objectID":"41448439","points":314,"story_id":41448439,"story_text":"Hi HN - we\u2019re the founders of Ligo Biosciences and are excited to share an open-source implementation of AlphaFold3, the frontier model for protein structure prediction.
Google DeepMind and their new startup Isomorphic Labs, are expanding into drug discovery. They developed AlphaFold3 as their model to accelerate drug discovery and create demand from big pharma. They already signed Novartis and Eli Lilly for $3 billion - Google\u2019s becoming a pharma company! (https://www.isomorphiclabs.com/articles/isomorphic-labs-kick...)
AlphaFold3 is a biomolecular structure prediction model that can do three main things: (1) Predict the structure of proteins; (2) Predict the structure of drug-protein interactions; (3) Predict nucleic acid - protein complex structure.
AlphaFold3 is incredibly important for science because it vastly accelerates the mapping of protein structures. It takes one PhD student their entire PhD to do one structure. With AlphaFold3, you get a prediction in minutes on par with experimental accuracy.
There\u2019s just one problem: when DeepMind published AlphaFold3 in May (https://www.nature.com/articles/s41586-024-07487-w), there was no code. This \nbrought up questions about reproducibility (https://www.nature.com/articles/d41586-024-01463-0) as well as complaints from the scientific community (https://undark.org/2024/06/06/opinion-alphafold-3-open-sourc...).
AlphaFold3 is a fundamental advance in structure modeling technology that the entire biotech industry deserves to be able to reap the benefits from. Its applications are vast, including:
- CRISPR gene editing technologies, where scientists can see exactly how the DNA interacts with the scissor Cas protein;
- Cancer research - predicting how a potential drug binds to the cancer target. One of the highlights in DeepMind\u2019s paper is the prediction of a clinical KRAS inhibitor in complex with its target.
- Antibody / nanobody to target predictions. AlphaFold3 improves accuracy on this class of molecules 2 fold compared to the next best tool.
Unfortunately, no companies can use it since it is under a non-commercial license!
Today we are releasing the full model trained on single chain proteins (capability 1 above), with the other two capabilities to be trained and released soon. We also include the training code. Weights will be released once training and benchmarking is complete. We wanted this to be truly open source so we used the Apache 2.0 license.
Deepmind published the full structure of the model, along with each components\u2019 pseudocode in their paper. We translated this fully into PyTorch, which required more reverse engineering than we thought!
When building the initial version, we discovered multiple issues in DeepMind\u2019s paper that would interfere with the training - we think the deep learning community might find these especially interesting. (Diffusion folks, we would love feedback on this!) These include:
- MSE loss scaling differs from Karras et al. (2022). The weighting provided in the paper does not downweigh the loss at high noise levels.
- Omission of residual layers in the paper - we add these back and see benefits in gradient flow and convergence. Anyone have any idea why Deepmind may have omitted the residual connections in the DiT blocks?
- The MSA module, in its current form, has dead layers. The last pair weighted averaging and transition layers cannot contribute to the pair representation, hence no grads. We swap the order to the one in the ExtraMsaStack in AlphaFold2. An alternative solution would be to use weight sharing, but whether this is done is ambiguous in the paper.
More about those issues here: https://github.com/Ligo-Biosciences/AlphaFold3
How this came about: we are building Ligo (YC S24), where we are using ideas from AlphaFold3 for enzyme design. We thought open sourcing it was a nice side quest to benefit the community.
For those on Twitter, there was a good thread a few days ago that has more information: \nhttps://twitter.com/ArdaGoreci/status/1830744265007480934.
A few shoutouts: \nA huge thanks to OpenFold for pioneering the previous open source implementation of AlphaFold\nWe did a lot of our early prototyping with proteinFlow developed by Lisa at AdaptyvBio we also look forward to partnering with them to bring you the next versions!\nWe are also partnering with Basecamp Research to supply this model with the best sequence data known to science. \nMatthew Clark (https://batisio.co.uk) for his amazing animations!
We\u2019re around to answer questions and look forward to hearing from you!","title":"Show HN: An open-source implementation of AlphaFold3","updated_at":"2024-10-14T00:18:58Z","url":"https://github.com/Ligo-Biosciences/AlphaFold3"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"24pfilms"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["basecamp","research"],"value":"I have a 3Dgame project starting up. I had used basecamp awhile ago and was very excited to hear they had a new rev. \nSadly, I was very disappointed to see that there is still no proper integration for artwork.\nI think there is a great opportunity for a startup to create a basecamp like product for artists and designers, with image roll-over, audio/video support, and vector art. And also have it link out to Github.\nMaybe there is a product but I sure haven't seen it in my research. If anyone has suggestions it would be greatly appreciated."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["basecamp"],"value":"Ask HN: Basecamp for Art and Code"},"url":{"matchLevel":"none","matchedWords":[],"value":""}},"_tags":["story","author_24pfilms","story_3920967","ask_hn"],"author":"24pfilms","created_at":"2012-05-02T20:23:59Z","created_at_i":1335990239,"num_comments":0,"objectID":"3920967","points":1,"story_id":3920967,"story_text":"I have a 3Dgame project starting up. I had used basecamp awhile ago and was very excited to hear they had a new rev. \nSadly, I was very disappointed to see that there is still no proper integration for artwork.\nI think there is a great opportunity for a startup to create a basecamp like product for artists and designers, with image roll-over, audio/video support, and vector art. And also have it link out to Github.\nMaybe there is a product but I sure haven't seen it in my research. If anyone has suggestions it would be greatly appreciated.","title":"Ask HN: Basecamp for Art and Code","updated_at":"2024-09-19T18:32:41Z","url":""},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"taubek"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["research"],"value":"UCF Researchers Create Lunar Regolith Bricks"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["basecamp","research"],"value":"https://www.ucf.edu/news/ucf-researchers-create-lunar-regolith-bricks-that-could-be-used-to-construct-artemis-base-camp/"}},"_tags":["story","author_taubek","story_33368007"],"author":"taubek","created_at":"2022-10-28T05:27:24Z","created_at_i":1666934844,"num_comments":0,"objectID":"33368007","points":1,"story_id":33368007,"title":"UCF Researchers Create Lunar Regolith Bricks","updated_at":"2024-09-20T12:26:28Z","url":"https://www.ucf.edu/news/ucf-researchers-create-lunar-regolith-bricks-that-could-be-used-to-construct-artemis-base-camp/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mettamage"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["basecamp","research"],"value":"I sometimes interviewed for US companies that stated they will hire an accepted candidate as a remote contractor first, then get the H1B visa process started and relocate within 6 to 24 months.
Is this legal?
If so, why don\u2019t more US companies do this? I\u2019ve heard that companies with about 20 to 40 employees do this [1].
Edit: this blew up, which is really cool! I want to note that it might be wise to view the replies in this thread as inspiration for your own research and not as legal advice.
[1] The hiring managers told me themselves. And Basecamp even states it publicly on their latest job ad.
P.S. If you're in the US/EU/World education space, checkout my profile :)"},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Is it legal for US companies to hire remote non-US contractors?"}},"_tags":["story","author_mettamage","story_24168590","ask_hn"],"author":"mettamage","children":[24168658,24168663,24168674,24168679,24168706,24168712,24168747,24168770,24168814,24168899,24168943,24168958,24168997,24169100,24169105,24169116,24169217],"created_at":"2020-08-15T13:09:24Z","created_at_i":1597496964,"num_comments":62,"objectID":"24168590","points":56,"story_id":24168590,"story_text":"I sometimes interviewed for US companies that stated they will hire an accepted candidate as a remote contractor first, then get the H1B visa process started and relocate within 6 to 24 months.
Is this legal?
If so, why don\u2019t more US companies do this? I\u2019ve heard that companies with about 20 to 40 employees do this [1].
Edit: this blew up, which is really cool! I want to note that it might be wise to view the replies in this thread as inspiration for your own research and not as legal advice.
[1] The hiring managers told me themselves. And Basecamp even states it publicly on their latest job ad.
P.S. If you're in the US/EU/World education space, checkout my profile :)","title":"Ask HN: Is it legal for US companies to hire remote non-US contractors?","updated_at":"2024-09-20T06:48:47Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"thikari"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["basecamp","research"],"value":"Hey folks,
i am working currently on https://sponsorgap.com/ and just finished a first report about companies sponsoring newsletters, websites or running ads for certain keywords.
Would love to get some feedback about the reports.
Sponsorgap Report - #0001
Education Market
Newsletter and Podcast
Screencast is sponsoring the Cult of Pedagogy Podcast.\nSkillshare is sponsoring the Morning Brew Newsletter.\nEdmodo is sponsoring the EdSurge Next Newsletter.
Keyword Research
Udemy is running ads for the keyword Online learning\nCoursera is running ads for the keyword Learn Python
Software Market
Newsletter and Podcast
DataDog is sponsoring a couple of Newsletters like JS Weekly and Hacker Newsletter\nBasecamp is sponsoring the Dense Discovery Newsletter
Website Ad
Fastly is running ads on Dev.to\nKlaviyo is running ads on StarterStory\nPusher is running ads on ProductHunt
\ufe0fRemote work
Newsletter
SafetyWing is sponsoring Nomadlist and Remote Health
ShieldGeo is sponsoring NoHQ Newsletter
Panther is sponsoring the Remotive Newsletter
Thanks for reading it! Any thoughts? How can I improve the report? What I should add?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Discover who's buying sponsoring and running ads in newsletters and websites"}},"_tags":["story","author_thikari","story_24687825","ask_hn"],"author":"thikari","children":[24688031],"created_at":"2020-10-05T14:02:33Z","created_at_i":1601906553,"num_comments":2,"objectID":"24687825","points":5,"story_id":24687825,"story_text":"Hey folks,
i am working currently on https://sponsorgap.com/ and just finished a first report about companies sponsoring newsletters, websites or running ads for certain keywords.
Would love to get some feedback about the reports.
Sponsorgap Report - #0001
Education Market
Newsletter and Podcast
Screencast is sponsoring the Cult of Pedagogy Podcast.\nSkillshare is sponsoring the Morning Brew Newsletter.\nEdmodo is sponsoring the EdSurge Next Newsletter.
Keyword Research
Udemy is running ads for the keyword Online learning\nCoursera is running ads for the keyword Learn Python
Software Market
Newsletter and Podcast
DataDog is sponsoring a couple of Newsletters like JS Weekly and Hacker Newsletter\nBasecamp is sponsoring the Dense Discovery Newsletter
Website Ad
Fastly is running ads on Dev.to\nKlaviyo is running ads on StarterStory\nPusher is running ads on ProductHunt
\ufe0fRemote work
Newsletter
SafetyWing is sponsoring Nomadlist and Remote Health
ShieldGeo is sponsoring NoHQ Newsletter
Panther is sponsoring the Remotive Newsletter
Thanks for reading it! Any thoughts? How can I improve the report? What I should add?","title":"Discover who's buying sponsoring and running ads in newsletters and websites","updated_at":"2024-09-20T07:10:02Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"meerita"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["basecamp","research"],"value":"I'm tired of seeing the same pricing systems on every startup it comes out: plans. I understand they are designed well and is common in the industry at this time but I think we can improve more. Some pricing plans are well distributed but some others, not. They just list a bunch of stupid features just to fill up the list and become something important to show.
I am developing a SaSS system and I want to implement a different approach for pricing. It starts with $10 for everyone, but you need to pay extra when you need to use or do other stuff.
Let's say you're 37signals Basecamp and, one feature is SMS alert. You either can pay the pro account extra 200 a year or instead you pay those SMS you need to send that x month for that x project you're working on. If it becomes a must feature, you pay a fee and you have it for default in your account. If not, you pay per use.
I've researched a lot for my project and I found people doesn't need all the features all the time. Most of the features I've been researching to put on the system are, more used occasionally than usually. I want to charge for the occasional use and, if it becomes a must, you can pay it too instead paying for something you will never use.
Has anyone tried this system and can share results? I don't know the side effects of this in real world projects. The only effect i tested with people told me that plans makes things attractive because \"it has that and that just in case i need it to use\" and that's all."},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: does pay per use pricing method works?"},"url":{"matchLevel":"none","matchedWords":[],"value":""}},"_tags":["story","author_meerita","story_5436251","ask_hn"],"author":"meerita","children":[5436335,5436505,5441317],"created_at":"2013-03-25T09:57:20Z","created_at_i":1364205440,"num_comments":5,"objectID":"5436251","points":3,"story_id":5436251,"story_text":"I'm tired of seeing the same pricing systems on every startup it comes out: plans. I understand they are designed well and is common in the industry at this time but I think we can improve more. Some pricing plans are well distributed but some others, not. They just list a bunch of stupid features just to fill up the list and become something important to show.
I am developing a SaSS system and I want to implement a different approach for pricing. It starts with $10 for everyone, but you need to pay extra when you need to use or do other stuff.
Let's say you're 37signals Basecamp and, one feature is SMS alert. You either can pay the pro account extra 200 a year or instead you pay those SMS you need to send that x month for that x project you're working on. If it becomes a must feature, you pay a fee and you have it for default in your account. If not, you pay per use.
I've researched a lot for my project and I found people doesn't need all the features all the time. Most of the features I've been researching to put on the system are, more used occasionally than usually. I want to charge for the occasional use and, if it becomes a must, you can pay it too instead paying for something you will never use.
Has anyone tried this system and can share results? I don't know the side effects of this in real world projects. The only effect i tested with people told me that plans makes things attractive because \"it has that and that just in case i need it to use\" and that's all.","title":"Ask HN: does pay per use pricing method works?","updated_at":"2024-09-19T19:26:36Z","url":""},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"spikey_sanju"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["basecamp","research"],"value":"What makes a great to-do app? Is it about features or a whole new way of using it? Could a different mindset be the key? Maybe a simple design?
I've tried all the famous apps: todoist, basecamp, linear, trello, asana, obsidian, and even github. But something feels missing. The process of researching, note-taking, logging, tracking, and completing tasks\u2014it's not quite right.
Imagine you're making a to-do app from scratch. What would you focus on? What new ways of thinking or methods would you bring in? What could you leave out?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: What Makes a Great To-Do App? Your Ideas?"}},"_tags":["story","author_spikey_sanju","story_37256920","ask_hn"],"author":"spikey_sanju","children":[37257022,37258124,37258132,37258435],"created_at":"2023-08-25T01:49:53Z","created_at_i":1692928193,"num_comments":4,"objectID":"37256920","points":2,"story_id":37256920,"story_text":"What makes a great to-do app? Is it about features or a whole new way of using it? Could a different mindset be the key? Maybe a simple design?
I've tried all the famous apps: todoist, basecamp, linear, trello, asana, obsidian, and even github. But something feels missing. The process of researching, note-taking, logging, tracking, and completing tasks\u2014it's not quite right.
Imagine you're making a to-do app from scratch. What would you focus on? What new ways of thinking or methods would you bring in? What could you leave out?","title":"Ask HN: What Makes a Great To-Do App? Your Ideas?","updated_at":"2024-09-20T15:00:46Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"abl"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["basecamp","research"],"value":"Ok, so its a little early, but I am very excited about the potential of Google Sites. As I was researching various wiki solutions for a corporate project collab site, I ended up settling on Google sites as a hosted wikifarm solution. The presentation is in google's repertoire - clean, lacking features, but very promising. :)
After creating a few sample pages and looking for more information, I found the Google Sites blog:
http://googlesitesblog.blogspot.com/
Even though Google Sites has been out for a few months now, based on the very fresh blog entries and a spurt of activity, I have a feeling google is ramping this app up.
If you examine the list features, specifically the project tracking templates, it almost seems like they are going after Basecamp as well.
Now, consider the power of embedding google gadgets, and my prediction may not be very far off!
I just don't understand why google doesn't add massive features to its apps products, given that they have a whole army of developers..."},"title":{"matchLevel":"none","matchedWords":[],"value":"Prediction: Google Sites - Ning Killer?"},"url":{"matchLevel":"none","matchedWords":[],"value":""}},"_tags":["story","author_abl","story_206389","ask_hn"],"author":"abl","children":[206487],"created_at":"2008-06-02T02:44:12Z","created_at_i":1212374652,"num_comments":2,"objectID":"206389","points":1,"story_id":206389,"story_text":"Ok, so its a little early, but I am very excited about the potential of Google Sites. As I was researching various wiki solutions for a corporate project collab site, I ended up settling on Google sites as a hosted wikifarm solution. The presentation is in google's repertoire - clean, lacking features, but very promising. :)
After creating a few sample pages and looking for more information, I found the Google Sites blog:
http://googlesitesblog.blogspot.com/
Even though Google Sites has been out for a few months now, based on the very fresh blog entries and a spurt of activity, I have a feeling google is ramping this app up.
If you examine the list features, specifically the project tracking templates, it almost seems like they are going after Basecamp as well.
Now, consider the power of embedding google gadgets, and my prediction may not be very far off!
I just don't understand why google doesn't add massive features to its apps products, given that they have a whole army of developers...","title":"Prediction: Google Sites - Ning Killer?","updated_at":"2024-09-19T16:18:39Z","url":""}],"hitsPerPage":50,"nbHits":9,"nbPages":1,"page":0,"params":"query=Basecamp+Research&tags=story&hitsPerPage=50&advancedSyntax=true&analyticsTags=backend","processingTimeMS":12,"processingTimingsMS":{"_request":{"queue":1,"roundTrip":14},"fetch":{"query":8,"scanning":3,"total":12},"total":13},"query":"Basecamp Research","serverTimeMS":15}