Alphafold is Amazing ! Here is a Tutorial
On Saturday, I ran Alphafold for the first time. Google built a website (Alphafold Server) through which you can run the program directly. The results are really positive, or cross that, extremely positive (wink). Here I provide a brief tutorial and a description of what I found.
If you are not familiar with Alphafold, it is a computer program that finally solved the long-standing and extremely difficult protein-folding problem. More recently, Google Deepmind solved all genetics problems as well. Regarding Alphafold, Nobel committee wrote -
Demis Hassabis and John Jumper have developed an AI model to solve a 50-year-old problem: predicting proteins’ complex structures.
In 2020, Demis Hassabis and John Jumper presented an AI model called AlphaFold2. With its help, they have been able to predict the structure of virtually all the 200 million proteins that researchers have identified.
Running Alphafold 3
My earlier blog post was based on results derived by three high-school students, and I realized that it was too harsh. After all, Jumper and Hassabis are Nobel prize-winning geniuses like Feynman, Dirac and Bardeen. Would you comment on Bardeen’s paper based on observations of high-school students? [1]
To test Alphafold, Here is what I did. I picked four proteins from the reference genome of Giardia duodenalis (GCF_000002435) somewhat randomly. I picked two for merely being near the top of the protein file from NCBI, and the other two from the list of hypothetical proteins. I also made sure that the proteins were long enough (335 aa, 745 aa, 1085 aa, 2015 aa) so that the folding program would work fine.
Here are the protein sequences if you like to run this analysis yourself -
MGLYNSSGTCTPCTDSCAVCKDGKSTGCQQCSPGKVLEPSVVANGNAKCVDECAVGDDCAECGITIDGSRYCTRCKDSNMYPLNGVCIPNTQKDAYCTNKANGACTTCSGAAFLMNGGCYTTEHYPGSTICDKQAGGKCEKTKKEYGISDDGKLLECDPTCLGCTAPGPGRCTRCPSDKLLKRASGAATGSCIDPNACVDGYYADGDACLPCATPGCKTCGHASFCTECAGELFVSLDGQSCLGECTGDKVVGEVPGGVRRCWCERGFLPALDRSECVLPTECPPDMPSCAACDTSGRCLLCAASNHNIQTDQRTCAEGCGARASSNQGVCHKNS
MERIVSPTGRVRKSPKRDPGSPRSAVRAVQHMTAVVSGIRAESQGATLTRGRLENVLLGRTASALSQSISSRAVSPLKKPPGSSLMSLIDANQPRPVMSNRRSPRPTRSARSRELGKCFANVTFIKTRAEGSNSMYYSTQSTLDESGELRSRPVPDLVEEINAVQLRSMHGEIQSPSESISETTNIQRTSLQNMSTKAECSFPADVQSKDSDRLSETDLVFIDPKKSFMQAQARRFIASEASPDRGLLYCSKAARALSVRGNIVPQIPFSKSSGTPGVSASAIRKQPPELFRPLSSSGRDSSFSRPRSTSTKIQPPRISSQPHIRSVSSDARTSIKPAARLVFPVPPKRHQLTPRERQALAHKRMRRETTWISDQLNGRALSFATLHRNASASRDASARRSLSAQKTSFCIFTGPQRIASPRVKPKQQWIPAQEACYTDDQDHTPKGRSGELRCSPHYVALSPTYAQPIAPSPIRNMDMTDVTLYSPNPSLVYPQRTGSIDEDLVNAASAKRTMVIEPRRATTATMEPNAMTLDRSLDIKLQGERPSKPMYTKAGQRIVPPQTIKPPARPARPIPPPNHNLQIYKRSPSNSSTTANQPLLRHVSKEVEQEAKHVVLEDLEDRHNSSTPQYVAPGSSNIISGVFDLRESALNRSSTKLDVVNLLPYFYYKQGETITLAQLPEIMKELNYNIDALKISAIKPQPHDQSSIFFQARVQTPQRRATSSGTTYERRRKRPKSASSPRMRI
MDEVTALRKRVVDLQVQSLGSSFTQEEITSIEDGIRFDAAQLLHRLFPYQSVKQHYNGSTICGDSGRFQANIAFITRLRAAILNAAEVQQAVKNGSFDRAESDLEGLGRLSPPEIDTKRYSCDSDDDRRSRSAKQKAQTLHVTISESEQELLRLQAEYASLNVVRQERDDYEEQELLQVEEALAAITRNSESALQQVMVSLAYAHALIEFTSLSSETKTEYIKLASTCDSEARRAVLGAEMSAVSDHVPDGRALASKVELVRTGLIKLFGEDVVDQGFTDLLAIQRERVIDAIADSVDNVLEAVMETWPRSHAKRETNTANAFESYDIPQDLGSLTTSDLKAPNMSRTVAADEVVAKPYVPETCKVRYPRLPIPSGDVGETDSVLSICVDEQNLRKQGPGSKKSDHSEAKACSTLESGSHVPEVSNISVLKPTDSPRLQIERAERERVTIYTKLRETITQRIEKGVRQTTPPDASPIEGPSVLSAIRNRASAPSTPGTQLEDVKNTTKVSHPPRHDSVDELISTVNKHMQRLESLTSSNTFSAVPAVQSPEIEVLFSRPQLDKISIDRRVSSTSHVSPEQLQADTPLRDTSELENVTEDYTFSQLQANDNDNDNNDEETSNNSKSYPTLTFDTAYDQDSLPQFLRGEDSSQKSIIAEKEETILRREEMRKQLEALYRQRELILQQKKQKTQQKETKKIDYRSERKLTPDITATEAKRKASQERIDKLATPRRRFDTPKRDEKNDEQSSQSSTTIAQSTRTPQKRTTSARSCTRETPKRDKQHVSFNSNSHTEARESTSKPSGISVGAYPSSIYLEETDELDVDSQSMTRPARSRPDSGGKSRLPRPILTDDLSLAAHNPRPPLPSYLESPPTVDLPGSKHSASTRTHSRHQTRSTTPPSTMPLSATYNFEHSGSYMPLRRERTLPHSSVIHPVIGPILSPLRSKILSETQFEIERKKAIRRRKEYERKEKERRAGETTVKEELLKQRDQRIRDFVKRMEGGNFQEKTFKKLLDKRSTGATSGQSGSLDQTTRARSKNRFDAEEEEQGAKRNNIDDRLTRASEAGYDCMDDGLKSILSDIEHDLFD
MNNHDSSCKDLSSVIADSETIEESIPNYVLLKYVLYEALKRVVEELPRKGQINPATRSILILSKIRTLLSFLKHLLPIYSTHSRDIISKGKQLILSRWASSHDHYRVLRSVDEFPVTFTEVHGFYDPIYAVDRAIVMLSGPPGTRLTRWPFYSTPISMLQASKERYKSLIHGQYKRATTEEVVRYMVRELKPAGMSLAVSGARVAVKTDFYRCVFVTDINFYLNSGIDFLPMELMKLQFYPAELAQNTIMISSITECIQEKLQRVRKSLEMTSIRPTDTRFYMVHSIICDFISSRAASTMVTQLAMYKASNQNSAFQARASFYRRAPNAYGVQLQDALHEIAVPSFHHFFDKHKHRLLQPCFAVIFRLTVWSNSLIEKVYELGGCMLEDVGLFYPHIDLKYMIGYSVDEDTSLTAVKSLALGVGKGSHQTIGHILTLYQQTKRKDNYAIFFFRPVYTYPFGLIQLLSDTRVKHRKGPDFAEYLDVTRIPYYSTDSTDLSTRSQSSVFLFRFIMSLHSQLSSHINIPLVDFYNHFKVILANNTFLVDATREGLQAEISSVEIVLDASLGLAVSIISPGGDGRLQLTFQIKGADVTKIVQRIFSAAIYSYTDTIVDFMSDFARYFFLISDLSAGNTEPHFSKLQFEELQAQPQANRYLVLEGIDPTTVFSLPPTDTTKSLYTEKVYKALRKFNYLLRTCRFSMEIPHTQYSLFGADIKSFKTNEGDSMNSLNSSLFMSIENISEQLNYYHGGYFIPTPIPHAHVAVAMWDKSVRFYSEFLKESMSVKSPITDMTIQQKAECCVFVALAYSLPNIISPSGLLSAKPSGAWAGSVIKSIPINSAYLKYLRCNPAYESDFLATYAFLNTTVTTCRVSGSAMCLSKLYGIVAHIDSLLWHRAIERLLLGYNPSVILESNKACSECKRFRQFIQRTSAHGDVSQESELTIQLFTLLLNRSPSYSALPYRLRIVYRTLPNSTEGAFSAKGSSVLNTMLIPQLRQCKACNCDLKNPCRSKAASALELHLILYTTPPDEANPDSTGEVERSIAIYNGRASYSLVSRALLTIAPTICCQLFAKAMIDNLPLHIYLQRSEKALQTQSSMQRQRATEVDFRAFTRRDLVLLLFNLRNYIMKTRELQFNYQSFISYILDPDVIPTNVKLNGCTPEQHKMFIKLIALHQDFVSKRSLRRCIYSFKLPGQKIPLIPFIPSQRLFGHQGAGSEHSLSASDHRVRELSYLNFMPGSQLKQQAKIWSDNKVSASQGLQESLVKVPGLPLEHYYNIVEEEEAIVIFTATSLAHLITQHSYRLDDAWSKTDNRSDKRSKPQQIYTHTLEFPGGYLNPEQLSCFLLREPSNKGLGPEHNKQTRQLLNCLFGRVYENARLLGVIFRMAYMPTTFFENRLLLEPLDTYATTEPHCIKLAGMEREFVATTTIEATDSLRARISTQRETFERAWPQSQESKYQPDKSATSHVKLTDSLSPMVVSGALHYKRITNTMAKQTQVDPLPGHTGLLHLSVRMLCYTNPSVLQNGPQSFYILFDLNPPSIYTQLTAVEEPSGYDYIGDTTSFPETVHLWLARSGGDGIPVFAPRHLRYKLAFDAPAFRYVKAYSNGVGHYLMRYYNNEVLPVISLGSLYKQFDEALVPYFNCNLIGSTSLDDQRRLVADIRMKKAFRVCLGLLDLSEKGDLLTLLRTFTLEKERMQAIAKNDQVSFSLIQSNGRIKPIFARMYARMHSLLKSYHDTLPTRQTVIAKTCNVLLNLLCGVTNTSSKLNQISPGVSVEDVLSIVMTVEIAPRYWLSDLCYESLDKELAMYLLQKYSRINLQYLSIHSLIMLTTINTVICQRALLQFMQHTFDAFLDRVSSKQIVSTNNAPYFINSSNISLLAAKSISVYGVSSFPEVTGSFKLEGRPVHNVDRARDFPPLCTILRTLQLTDIISRRSMRLELLSYSMLIVRDGAEKMTGQHQSVSLGMPQQDTRLFLPIYIDRDPSVNRNLLQYYKTIRYGKYVIIYSLDGKFIGRIVQ
The method for running Alphafold is quite straightforward. You will run Alphafold 3, which is a better version than the one described by the Nobel committee. You need to log into Alphafold 3 server with your google account, sign a contract saying you will use the software only for non-commercial purposes and you will donate your first child to Google if results are useful. Then Google will “generously” allow you to fold 30 proteins per day.
The method is quite simple. You copy and paste your sequence in the form, press the blue button saying “continue and preview job”, and your job will enter the queue. You can track the queue at the bottom of the same page, and then after a few seconds to minute, your result will be ready.
Here are the folds I got -
Let me explain how to interpret these results. The colors bluish to deep blue mean Alphafold is confident about the fold, and orange to red means Alphafold thinks the results are AI slops. The page also includes a metric called pTM, and provided this helpful description.
A pTM score above 0.5 means the overall predicted fold for the complex might be similar to the true structure. ipTM measures the accuracy of the predicted relative positions of the subunits within the complex. Values higher than 0.8 represent confident high-quality predictions, while values below 0.6 suggest likely a failed prediction. ipTM values between 0.6 and 0.8 are a gray zone where predictions could be correct or incorrect.
Summary in Plain English
From the posted images, intelligent readers may have already figured out, where this story is going. Therefore, this plain English summary is only for the members of Nobel committee and other idiots.
The image colors show that 3 out of 4 folds are AI slops. That means 75% of the output of this world-changing software are fake based on their own measures. Yes, Alphafold can fold 200 million proteins, but how many of those folds are real? The low quality of folds is also confirmed by the extremely low pTM scores for those proteins.
How about the remaining one? Even the one that looks mostly blue has a pTM of only 0.63, and thus it is in the “gray zone where predictions could be correct or incorrect”. Thus one can argue that all of the folds are c**p. This is not surprising, because the PDB database has only a handful of proteins from Giardia, and most of those are structures of ribosomal proteins. What is surprising that Alphafold claimed (and made Nobel committee repeat the claim) that it “solved 50 year old protein-folding problem).
My bigger concern is that even this determination of “confident” is based on some algorithm. Who ranked this algorithm, and how do we know that Alphafold’s confidence in the fold is real? Is Alphafold overly confident in its abilities just like the members of the Nobel committee are in theirs?
Please note that I ran Alphafold 3, a better version of the protein folding program than the one that “solved protein folding problem for good” according to the Nobel prize committee.
Why Alphafold is Still Amazing
I will have to speak in tounges and keep repeating that Alphafold is a great software and Jumper and Hassabis are smarter than Feynman and Dirac. It is because my earlier post titled “Alphafold, the Brawndo of Modern Science” got deranked by Google.
You can do the experiment yourself. Please copy and paste the title of any of my last 10 posts on Google. For all except one (and you can guess which one), Google will take you straight to the post. Coincidence?
[1] That was a tounge in cheek. I believe Brian Josephson discovered Josephson effect, when he was in high-school, and Bardeen thought he was wrong. Also, Josephson used to comment in this blog, when it had a comment section.
