(Fig.1) Deepmind Alphafold is unable to simulate protein conformational changes caused by mutations, interactions with other molecules, hence useless for drug development or curing cancers.
Many mainstream media claimed "Google's Deepmind or Alphafold AI published in the recent Nature journal could solve 50-year-old protein folding problem or predict some protein structures, huge breakthrough, making history !".
But it is completely untrue. Actually, Deepmind Alphafold has Not solved or clarified the true mechanism of protein folding at all, and many proteins's structures are still impossible to predict by Alphafold ( this 5th-last ~ 4th-last paragraphs ), hence they are useless for drug development or curing diseases.
All the current protein folding prediction tools are trained based on already-known experimentally-verified protein structures stored in protein data bank (= PDB ), they do Not use the (useless) quantum mechanical or molecular dynamical simulations for protein folding prediction.
Because the process of protein folding takes milliseconds ~seconds~hours ( this p.2 ), which time scale is completely out of reach of even the current fastest protein simulating methods (= molecular dynamics or MD ) that take too much time.
Even the current fastest protein simulating method = molecular dynamics (= MD ) takes more than days to simulate just nanoseconds~microseconds (= μs ) of motion or conformational change of a protein ( this 5th-paragraph, this 3rd-paragraph ).
↑ Such an unrealistically-time-consuming, impractical molecular dynamics cannot simulate the millisecond~second-long protein folding or interaction process ( this p.1-intro-2nd-paragraph, this p.2-left-middle ).
The fact that Deepmind Alphafold's prediction is based on the experimentally-obtained protein structures of PDB (= and they actually has Not find the real mechanism of protein folding ) means Alphafold can Not predict the structures of new proteins or mutated proteins at all.
This p.2-right says
"AlphaFold deploys deep-learning neural networks.. It has been trained on hundreds of thousands of experimentally determined protein structures and sequences in
the PDB (= protein data bank, this p.1-right, this p.9-right-Data availability ).."
".. The network’s reliance on information about related protein sequences means that AlphaFold has some limitations. It is Not designed to predict the effect of mutations, such as those that cause disease, on a protein's shape. Nor was it trained to determine how proteins change shape in the presence of other interacting proteins, or molecules such as drugs."
This p.6-last paragraph says
"our data indicate that AlphaFold predictions cannot be used directly to reliably estimate the impact of mutation on protein stability or function.. Indeed, AlphaFold was not designed to predict
the change of protein stability or function due to mutation ( this 1~2nd paragraphs )"
And Alphafold just outputs one (useless) static protein structure from the input amino acid sequence ( this 2nd-paragraph, this 10th-paragraph, this middle-Where AI falls short ).
Actual proteins are known to form various different conformations or folding structures even from the same amino acid sequences, depending on how they interact with other different molecules and proteins, which dynamical protein change (= important for actual biological reactions ) can Not be predicted by Alphafold ( this 9-10th paragraphs, this 2nd-last-paragraph ).
This 29~35th paragraphs say
"Indeed, AlphaFold and other similar self-didactic programmes can predict only the static 3D structure of a protein.. But proteins are very dynamic."
".. But none of the tools scientists can currently access are capable of generating a clear picture of how the protein structures change in time, and in response to chemical changes... So AlphaFold’s achievement is significant – but it hasn't exactly ‘solved’ the problem itself."
This 4th-paragraph says
"they still only offer (static) snapshots of proteins. To capture whole biological processes—and identify which lead to diseases—we need predictions of protein structures in multiple “poses” and, more importantly, how each of these poses changes a cell’s inner functions."
For example, as shown in the upper Fig.1, various different knot patterns (= corresponding to different protein folding patterns ) are the result of proteins choosing different folding processes ( of how proteins interacted with different surrounding molecules and proteins ) produced from the same rope (= corresponding to the same amino acid sequence ), which means the attempt to predict the single static protein structure only from the amino acid sequence as Alphafold tries to do is meaningless, and unable to clarify true protein folding mechanism.
To really understand the protein folding mechanism, we need to simulate dynamical conformational changes as the result of how proteins interacted with other molecules, proteins and enzymes in the protein-folding process (= like different rope's knot patterns made from different rope-tying processes ), which dynamical simulations take too much time, hence impossible to simulate by the current useless mainstream protein-simulating methods such as quantum mechanics and molecular dynamics.
Actually, recent experiments showed that Deepmind-Alphafold AI turned out to be useless for predicting protein interaction or dockings necessary for drug discovery.
This 10~16th paragraphs say
"AlphaFold was Not very effective for modelling molecular docking simulations accurately.. AlphaFold was not really trained for molecular docking simulations. Docking small molecules into a given protein structure is really a different problem than determining that protein structure."
"Being able to model these types of chemical interactions is an unsolved problem. No algorithm is perfect ( this abstract-lower, this last-paragraph )."
↑ In order to "really solve" protein folding problem, we need to understand and simulate protein-protein interaction in the process of protein folding, which dynamical simulation is out of reach of all the current protein folding prediction methods such as Alphafold.
And Alphafold can Not predict post-translational modifications of proteins and interactions with ions and ligands that are necessary for actual biological reactions, hence, Alphafold cannot be used for predicting effective drug binding sites ( this 7th-paragraph, this p.4-left 2nd-paragraph ).
This 3~5th paragraphs say
"AlphaFold deals with amino acid sequences, nothing else.
This means that the tool is Not capable of predicting metal ions, cofactors or ligands, although they are of great importance to the folding of many proteins (including haemoglobin, the oxygen-binding protein in red blood cells).."
".. Post-translational modifications (PTMs) of proteins, such as phosphorylation of glycosylation, can significantly impact protein structures. But again, AlphaFold deals with amino acids, nothing else – so PTMs are Not considered in AlphaFold’s structure predictions".
Even in the prediction of the already-known static protein structure, Alphafold is still impractical, as shown in the last paragraph of this latest research news grudgingly admitting,
"While still Not reliable enough to replace the slower methodical X-ray crystallography for structure or a controlled assay experiment for function.. "
(Fig.2) Contradictory quantum mechanical Pauli principle rule prohibits each atom from having definite shape and boundary. ← Impractical model forever.
As I said, in order to understand true protein folding mechanism and protein interactions with other molecules, it is necessary to simulate protein conformational changes using some realistic and less-time-consuming atomic models.
Protein folding and protein ligand-binding processes usually take milliseconds~seconds ~minutes, and protein synthesis by ribosomes takes much longer times = hours, which important biological reactions' time spans are far beyond the calculating capacity of the current mainstream protein simulating methods ( this p.14 3rd-paragraph ).
Even the current fastest protein simulating method called molecular dynamics (= MD ) takes more than days to simulate only nanoseconds~microseconds of protein's motion ( this 3rd-paragraph, this p.3-right-first-paragraph, this p.5-Figure.2 ).
The original quantum mechanical method takes much longer time, hence more impractical than molecular dynamics.
↑ This means even the current fastest molecular dynamics (= MD ) would take unrealistically too much time = many years to simulate even simple protein folding or milliseconds~seconds of protein conformational change, which simulations are impossible ( this 5th-paragraph, this p.2 2nd-paragraph, this p.2-left-first-paragraph, this p.1-intro-2nd-paragraph, this p.2-right-first-paragraph ).
The fact that all the current protein simulating methods are useless, taking unrealistically too much time for simulating important biological processes such as protein conformational change, folding and interactions means it is absolutely impossible to develop effective drug (= by considering protein interaction with drug molecules and conformational change ) or cure diseases, as long as we blindly rely on the current useless mainstream quantum mechanics and molecular dynamics.
The main problem of the current useless time-consuming quantum mechanics and molecular dynamical simulations originates from the quantum mehanical contradictory, Pauli exclusion principle model expressed as unphysical Pauli antisymmetric wavefunctions and exchange energy (= lacking real exchange force ).
It is known that only some limited places or bonds can be rotated or changed in polypeptide chains of proteins, which means the actual protein conformational change is very simple, and the simulation of such protein structual changes can be easily and quickly performed, if we treat each atom and amino-acid as a real object with tangible shape like ordinary parts composing practical macroscopic running machines.
In the upper Fig.2, only two bonds (= ψ, φ ) can be changed or rotated, and other parts such as (yellow) plane can be treated as rigid objects with shape and boundary that can be rotated until one of those atoms contained in the rotating rigid body collides with other external atoms and is stopped like ordinary rotating macroscopic rigid objects.
Like practical parts ( with their concrete shapes ) composing cars, planes, engines and machines, we can easily "simulate" motions of those real parts and atoms, if we know the concrete shapes and properties (= hardness, elasticity, stickiness.. ) of each tangible part or (realistic) atom.
The problem is quantum mechanical Pauli principle model does Not allow each atom or molecule to have definite shape or boundary, though the current atomic force microscopic technology can easily measure each atomic shape or boundary as Pauli repulsive contact force.
According to the unrealistic quantum mechanical Pauli principle expressed as unphysical Pauli antisymmetric wavefunctions, each single electron must always exist in all different atoms simultaneously to cause unphysical Pauli exchange energies.
For each atom to have its concrete shape and boundary (= electron's orbit gives shape and size of each atom ), each (indivisible) single electron must belong to only one atom at once, which atom consisting of a nucleus and separable electrons must be distinguishable and separable from other atoms. ← This is the realistic atomic and electron model.
↑ If each electron unrealistically spreads over all different atoms to satisfy the unreasonable quantum mechanical Pauli antisymmetric wavefunction rule or exchange energy, all different atoms also become unrealistically inseparable (= each atom with No boundary or No shape due to the spreading electron ) from other atoms due to one single inseparable electron always unrealistically bridging all different atoms.
In the real atomic model in the upper Fig.2, when the moving atom-A pushes the one side of the (rigid) rod consisting of three tightly (= covalently ) bonded atoms 1~3, this rod is eventually rotated around the fixed atom-B after the other side of the rod collides with the atom-B (= if this atom-B is fixed at other positions, ex. collides with the middle part of the rod, this rod's rotating pattern changes ).
↑ In this real atomic model that can give concrete tangible shape (= each atomic shape, boundary and contact force are experimentally-measurable by atomic force microscope ) to each atom and molecule, we can easily "predict" the molecular or protein conformational changes (= such as the rod-like molecular rotation above ) after colliding with other atoms like easily predicting motions of billiard balls with concrete spherical shapes after colliding with various other objects, even without performing extremely time-consuming molecular dynamical or quantum mechanical calculations.
On the other hand, in the contradictory quantum mechanical atomic model, each atom with No concrete shape and No boundaries is unrealistically indistinguishable and inseparable from other atoms, as shown in the current mainstream density functional theory (= DFT ) treating all different atomic electrons as one inseparable pseudo-electron.
Of course, we can Not define real (contact) forces between these unrealistically inseparable quantum mechanical atoms, as seen in the fact that quantum mechanics cannot treat experimentally-measurable Pauli repulsive contact force (= exchange force ) as real force ( this p.10, this p.4-middle, this p.3 3rd-paragraph ).
So in the current mainstream protein simulating models such as molecular dynamics, physicists have to artificially prepare pseudo-potential energies (= based on unphysical Pauli exchange energy or pseudo-kinetic energy composing artificial Lennard-Jones potential ) called force field (= denoted as U, V or E ) in each atom ( this p.3-6 ), by artificially choosing and adjusting the pseudo-potential parameters.
And by differentiating each pseudo-potential U acting on each atom, they try to obtain fictitious (exchange) force (= F = ∇U or ∇V ) acting on each atom (= contradicting the fact that quantum mechanics does Not admit Pauli exchange force to be real force ), update each atomic position and velocity one by one, little by little at each time step repeatedly many, many times, which calculation takes an enormous amount of time ( this p.5-24, this p.5-9 ).
↑ The time interval between time steps must be extremely short (= only 2 femtosecond, 2fs = 2 × 10-15 seconds ), which means the simulation of only a microsecond (= 10-6 ) motion of protein needs as many as 109 iterative time steps and calculations that take too much time.
If physicists choose the time intervals longer than 2fs (= which means each atom can roughly move a longer distance in one parallel movement ) to save time, it increases the chance that two atoms unrealistically overlap each other (= because quantum mechanical atom has No shape or No boundary, the time when two shapeless atoms collide with each other is unpredictable and unknown in advance ), and the total energy unrealistically increases (= called energy "exploding" ) due to the drastically-increased Pauli repulsive energy caused by the unrealistically-overlapping atoms, which eventually violate total energy conservation law and give wrong protein conformational results ( this p.9-10, this p.1-last~p.2 ).
As shown here, as the unreasonable quantum mechanical Pauli antisymmetric wavefunction rule prohibits each atom from having concrete shape or boundary, physicists can Not easily predict the molecular or protein conformational change such as rotation and collision between shapeless molecules. Physicists have to perform the extremely-time-consuming molecular dynamical simulation by differentiating artificially prepared pseudo-potential or force field at short time interavels, repeatedly many, many times for gradually moving atoms one by one, little by little.
↑ These current impractical mainstream protein simulating methods based on the unreasonable quantum mechanical Pauli antisymmetric wavefunction rule are contradictory and double-standard, hence wrong.
Because when scientists and engineers use ordinary macroscopic objects or parts for building cars, planes, machines and laptops, they automatically define and give concrete shapes to each part and component, adopting real contact forces without conducting the ridiculously time-consuming molecular dynamical or quantum mechanical simulations.
And only in the microscopic world, they suddenly stop treating each atom as a real tangible object by stopping giving concrete shape and boundary to each atom, though each atomic shape and boundary are experimentally measurable by atomic force microscopes.
↑ Both contact forces between macroscopic objects and between microscopic atoms are supposed to be caused by (unphysical) Pauli exchange repulsive energy, but scientists refuse to use this stupid Pauli exchange energy or antisymmetric wavefunction only when they build practical cars, planes and machines in macroscopic world, while in microscopic world, they continue to use this unphysical Pauli exchange energy and extremely-time-consuming molecular dynamical simulation method stopping scientific progress. ← Double standard !
Solution to this current dead-end basic physics is simple, as you may notice.
We just give concrete shape and boundary to each atom and molecule measured and confirmed by the current atomic force miroscopy technology, like macroscopic practical objects with shapes.
↑ We can easily predict and design various motions of a little bigger molecular device by freely combining several molecules and atoms whose shapes are already confirmed and known.
↑ Then, we can confirm this molecular device's motion , comparing it with the quick molecular motion's simulation based on real atoms with known measured concrete shapes, and adjust each atomic shape and property's parameters to experimental observation. ← By repeating this process, we can easily build arbitrarily-bigger molecular or protein devices (= like building small~large cars, planes and macroscopic machines ) whose motions can be easily and precisely predicted and controlled, hense useful for drug developing and curing diseases.
Unrealistic quantum mechanical atomic model cannot carry out this process indispensable for building useful molecular, protein or macroscopic devices, because physicists cannot give concrete shape to each atom or molecule (= hence, quantum mechanical model is unable to predict molecular or protein motions by predicting when and how shapeless atoms collide with each other ) due to unreasonable Pauli antisymmetric wavefunction or exchange energy rule.
Physicists have to blindly and artificially adjust interaction parameters of the pseudo-potential, Pauli repulsive energy or force field ( this p.8-9 ) without knowing each atomic shape, and perform the extremely-time-consuming, impractical molecular dynamical simulation with No ability to predict the final molecular conformation.
Hand manipulation of atoms and molecules by atomic force microscopes (= AFM ) usually takes more than seconds (= for practical purpose), which time scale is out of reach of the impractical molecular dynamical or quantum mechanical simulations ( this p.1-left ).
As a result, in order to clarify true protein folding and interaction mechanism for developing medical technology or building useful nano-machines, we just need to treat each atom and molecule as real object with concrete shape like the practical macroscopic parts for building practical cars, planes and macroscopic machines.
As you know, ordinary macroscopic tangible objects have definite shapes, hence microscopic atoms composing the macroscopic objects also must have their definite shapes. ← The unphysical quantum mechanical Pauli principle rule ignores this obvious fact !
But in order to admit each atomic shape realistically, we need to admit the existence of some real objects (= the same as the space medium causing particle's interference ) as the source of real Pauli repulsive contact force, which real Pauli contact force contradicts not only (unphysical) quantum mechanical Pauli antisymmetric rule but also Einstein relativity rejecting any real medium in the space by relying on unreal virtual particles.
↑ To protect the vested interests around the unrealistic dead-end 100-year-old quantum mechanics and Einstein relativity, academia, universities and "scientists" are forced to give up reality, refuse to progress really useful technology of curing diseases, and artificially create fictitious harmful scientific targets such as ineffective vaccines, global warming and parallel-world quantum computers.
2023/6/2 updated. Feel free to link to this site.
