Principles of Protein Structures

• Group research: Design of protein molecules for novel functional and behavioural activities. Also study proteins using variety of techniques like ML and computational design.

Moplecular Interactions in Biomolecules

• Non-covalent interactions modelled via Lennard-Jones, consists of van der Waals interactions, hydrogen bonds.
• Two neutral atoms in close proximity get polarised !
• Van der waals radius is r1 + r2 and is the point where the repulsive interactions become dominent.
• Hydrogen bonds can be thought of as a dipole-dipole interaction but on a molecular scale.
• Iconic interactions are the strongest, however, water ions severly reduce electrostatic interaction.
• In vacuum its ~50 kJ/mol, whereas in water its ~6 kJ/mol.

Moleculaes of Life

1. Nucleic Acids
2. Proteins : Made of Peptides
3. Lipids : Fats
4. Glycans: Sugars

• Proteins have an amino group, a carboxyl group, and 20 different kinds of R groups.

• They are built out of 20 amino acids encoded in DNA.

• The amino acids can be classified based on the type of Carboxyl group.

  • Polar and non-polar amino acids. Polar amino acids can form hydrogen bonds with itself and other amino acids. They are also hydrophobic.
  • Negatively and positively charged amino acids. Their sidechains are charged at pH=7.

• Synthesis of protein: condensation reaction where amino acid of one …TODO

• The sequence of a protein is written from the N-terminus (one of the free $NH_3$) to the C-terminius (the one with $COO^-$).

• backbobne + sidechain + actual witten structure + carboxyl extension.

• Hydrophobicity of amino acids is an important feature.

• Residue: the central kernal of the proteins. Find out why a Residue is called a residue. TODO

• Some amino acids are special:

  • Cysteine can form disulphide bridges.

    flowchart LR Cysteine -."+".- Cysteine2(Cysteine) --Oxidation--> Cystine Cystine --Reduction--> Cysteine2

  • Glycine doesn’t have a sidechain group.

  • Proline’s sidechain group is covalently bound to the nitrogen of the peptide bond.

Protein Secondary Structures

• $\alpha$-helix form to stabilize the hydrophobic core. It forms by making H-bonds between Hydrogen and oxygen.

• $\beta$-sheets form hydrogen bonds with its whole backbone.

• Usually the alpha and beta sheets are sometimes misaligned and imperfect.

• Alpha-helix

  • In the alpha-helix, the carbonyl oxygen of residue “i” forms a hydrogen bond with the amide of the residue “i+4”.
  • There are only right handed alpha-helixes, except for a handful of exceptions.

• Beta-sheets

  • Can be wither parallel and antiparallel. They have roughly the same binding energy.
  • In a beta-sheet, carbonyl oxygen and amides form hydrogen bonds between the strands, i.e between amino acids far away from each other in primary sequence.

• Proteins can be composed of one of the two secondary structures or be mixed.

Beyond Secondary Structures

• The tertiary structure, which is the three-dimensional organisation of the secondary structure elements. It is also referred to as the protein fold.
• Quaternary structures refers to the association of different polypeptide chains (subunits) into a multinumeric complex.
• Protein domains are fundamental units of tertiary structue. it forms an independnt structural domin. They are often units of function.
• Domains have the same fold yet different amino acid sequences.
• Protein domains are very stable to mutations. ANd they have been selected via evolution for the very specific reason.
• The stability of the folded structure results primarily from the hydrophobic regions folding in together.
• Two proteins with different sequences can have the same sequence.

Conformations and Folding

• Protein conformational changes does not require breaking bonds.
• The peptide bond has partial double bond character and therefore are sterically hindered.
• Protein folding induces conformational changes in the backbone. The backbone torsion angles $\phi$ and $\psi$ determine that conformation of the protein chain.
• The Ramachandran diagram define the restrictions on backbone conformation.
• $\psi$ : TODO
• $\phi$ : TODO
• L: Left handed helix
• You see a lot of Glycines in loops as they are flexible because of the absence of any side chains.
• Prolines are in Cis conformation so they ususally “kink” the alpha-helix. Not generally found anywhere else.
• Secondary structural elements are conected to form simple motifs. Given motifs, you can create many folds.
• The interactions between the atoms in a protein control the folding of the protein into a well defined structure (native structure): Thermodynamic hyposthesis of protein folding. It was demonstrated by the famous Anfinsen Experiment which opened the folded structure of a protein and allowed it to relax in a controlled manner to recover the folded structure,
• This information is encoded in their amino acid sequence.

Evolution and Protein Structure

• Reading Suggestion: The Molecules of Life (Chapter 5)

• Protein structures are conserved during evolution while amino acids vary.

• Motifs that are crutial for protein folding are conserved during evolution.

• You can use protein globins to classify species at the molecular level.

• Due to different selection pressures amd line seperation, protein sequences diverge between species.

How to quantify differences in sequences?

• Use a substitution matrix (BLOSUM). We use a matrix of probailities. Each entry in the matrix is called a substitution score ($S_{ij}$). The higher the score (positive), the more likely is the substition.

• The BLOSUM substitution score is the logarithm of the likelihood that one aa in a protein is replaced by another during evolution.

• $S_{ij}$ is related to the frequency of substititon over random chance.

• BLAST (Basic Local Alignmnet Search Tool) : Identify novel proteins, find functional relationships and sequence motifs and patterns.

• TOP-7 is the first synthetic motif.

• The most used measure is the Root Mean Square Deviation (RMSD).

• Domains are “self contained units” that can be placed in different proteins.

PyMol Excercises

• Superimposition is done by fitting the polypeptide backbone of your target protein on the reference protein. Best fit will have smallest Rms (root mean square) value.

• What is PSIPRED?

• Protein crystals have a high solvent content, which you can also observe as large spaces between the proteins in the Electron density maps.

Day 2

X-ray Crystallography

• The further out spots the higher the resolution of the protein structure.

• The dark ring ~3Å is caused by water molecules un the crystal.

• flowchart LR crystals --"x-rays"--> diffraction-pattern --> electron-density-maps-->atomic-structure atomic-structure --refinement-->diffraction-pattern

• Methods to check crystallization conditions for proteins: Hanging Drop- and Sitting Drop- Vapour Diffusion.

• Typical concentration for X-ray diffraction: TODO

• High B-factor: low confidence in position of atom, which also corresponds to low electron densities.

• In 2020 the number of structures in PDB reached 171,916.

Nuclear Magnetic Resonance (NMR)

• Resonance Frequency in NMR: $$ R_f = \frac{2 \mu_BB_o}{h} $$

• Resonance frequency depends on nature of the nucleus and its chemical environment.

• Resonance frequency is directly proportional to the strength of the external applied magnetic field.

• We commonly use $H^1$ and $C^{13}$ for labelling as they are NMR active.

• Nuclear Overhauser Effect (NOE): Provides information on spatial proximity of protons in protein. Protons that are close by influence each others chemical shifts.

• Crystallography si also applicable to very big proteins and complexes.

• NMR: Dynamic ensemble of several energy-minimized structures, whereas, X-ray crystallography gives a static picture of protein.

• NMR has a size limit: >25kDa is very difficult.

Cryo-EM

• Cryo-EM prevents electron degradation of proteins.

• Abbe’s Relationship: $d_R = 1.22 \lambda /NA$. Working out this formula gives us a $d_R$ to be 1.4Å.

  flowchart LR biochemical-preperation --> cryo-em-sample-preparation --> Imaging --> data-collection(data-collection) dc(data-collection)-->image-processing-->reconstruction--> structural-analysis--> Model

• Generally Liquid Ethane is used.

• 3D images are reconstructed from 2D segment images from all angles.

• Cryo-EM can do very big protein structures.

Bioinformatics and protein structure

• Uniprot : Entries are manually annotated. This is the first stop to check an unknown protein. It contains references to other databases as well.
• There are multiple .pdb files for a single protein corresponsiong to all the different experiments done with the protein. They might have different resolutions, and precision.
• PhosphoSite Plus : provides comprehensive information and tools for the study of protein post-translational modifications (PTMs) including phosphorylation, acetylation, and more.
• STRING: All about protein interations. It contains protein-protein interaction data.
• PDB: The Protein Data Bank.
• Drugbank: Database of drugs.
• ChEMBL: Drug database where searches can also be compound-centric using different formats.

AlphaFold

Each step involves shrinking of the total phase space of that particular function.

• Critical Assessment of protein Structure Prediction (CASP) - competition held for prediction of Protein Structure. The FOM is GDT (Global Distance Test).
• AlphaFold 1 was the first to combine Multiple Sequence Alignment (MSA) and Deep Neural Networks.
• MSA finds similar sequence across different species, aligns, and then find “suitable mutations” or the set of all possible mutations (coevolution).
  • If something mutates to a more positive base, the other base can mutate to something more negative, but not positive.
  • We calculate distance matrix -> Correlation matrix.
• A neural network can approximate any function, given that your neural network has arbitrary width and we have an infinite dataset.
• AlphaFold also uses a Convolutional neural networks to recognise patterns.
• Output is the distance matrix, and the distance matrix of the two dihedral angles $\phi$ and $\psi$.
• The Input for AlphaFold-2 is the MSA.

AlphaFold-1:

AlphaFold-2:

• Bad templates are inserted so AF won’t copy template features.

• AMBER relaxation optimizes the final structure.

• “Self-distillation training” - Using your own predictions as a dataset.

• Human vs Other Species -> Prediction Score Comparision

• Is all AlphaFold code open source? : Everything is opensource.

• Autocompletion is based on the Attention Matrix, whcih comes from a language model.

• AF also does a bunch of other things: protein-protein interactions, docking, predicting binding sites, etc.

• ColabFold is a better version of AlphaFold 2. It is a lot faster with improved MSA step. It also allows for complex predictions.

• AlphaFold 2 practice notebook

Day 3

Binding

• Spontaneous process at constant temperature and pressure always involves decrease in the Gibbs Free energy of the system ($dG < 0$).

• It accounts for both enthalpic and entropic contributions of the system. $$ dG = dH - TdS $$

• Equilibrium Constants:

$$ \Delta G^0 = -RTlnK_{eq} \rightarrow K_{eq} = e^{- \frac {\Delta G^0}{RT}} $$

• Homo Sapiens: ~15, 000 proteins expresses. 135,000 pp-interactions.
• Specificity is the ratio of binding affinities. They are theoreticlly not related. Innature, Higher affinity implies lower specificity.
• Binding sites usually have a hydrophobic core and then on the rim, you have bases that have specificity.
• Whwn fraction $f$ is half in a reaction, out dissociation constant $K_D$ equals the ligand concentration $L$.
• Univeral binding isotherm is expressed in terms of $K_D$.

Drug Binding by Proteins

1. Find a Lead compound. That bind to the receptor. They are usually not very specific and affinative.
2. Iterate the lead compound to imrove specificity and affinity.

• Competative inhibitors: $IC_{50}$ is the concentration of the inhibitor that reduces the binding of the protein by half maximal value.
• $IC_{50}s$ are used rather than a single value because the activity of the drug may vary between the population just because of random mutations.
• Energetics of the Binding: There is entropic loss but enthalpic gain. $\Delta G = \Delta H - T \Delta S$
• The affinity for a protein for a ligand is characterized by the dissociation constant, $K_D$.
• The strength of binding interactions is often correlated with hydrophobicity.
• MASIF: Surface-Based Protein-Protein Interaction Prediction
  • All informtion required for predicting binding is present on the surface.

Exploring the potential of deep learning in de novo protein design

• Protein perform a variety of functions

  • Signalling
  • Immunity
  • Structure
  • Catalysis
  • Transport

• Other functions (adapted for use):

  • Therapeutics
  • Biosensors
  • Enzymes
  • Biomaterials

• Proteins as domains:

  graph LR Sequence --> Structure --> Function

  As a protein enginner you reverse the graph in your approach.

• Motif grafting: Take parts of proteins (motifs) and combine them at the sequence level.

• Classic de novo protein design is limited by already known structures (the PDB database). The top-down design method

• A better approach is the de novo protein design: It takes a Bottom-up functional protein design. It is much more computationally expensive and suffers from a very low success rate.

• Added an MC Step to the protein iteration step for scoring random mutations.

• “Oracle models”

Exercise Notes

• “MaSIF receives a protein surface as an input and outputs a predicted score for the surface on the likelihood of being involved in a PPI.”

  — Assignment sheet

• MaSIF can be used for iteratively for negative designing i.e. to remove features from the protein structure.

Day 4

DNA structure and replication

• The famous Photo 51 which was an X-ray crystallographic image of the DNA double helix strand.
• DNA replication proceeds only in one direction. It is because of the presence of phosphtes on the one side only, which is required for catalysis.
• The replication movement creates a “replication fork”, which leads to a leading strand (continuous synthesis) and lagging strand (discontinuous sequence). The fragments on the lagging strand are vcalled Okazaki fragments.
• Replisome;
  • Average human chromosome: 150 Mbp
  • Average Eukaryotic replisome speed: ~50 nt/s
  • It will take 400 hours for the replication to occur.
  • Hence there are 100s of origins for replication on a chromosome.
• How to identify DNA replication origins? (which are parts ) :
  • Chop DNA, make some kind of loops with URA3. Only segments with Origins replicate and survive to grow into colonies.
• Deep Sequencing: sequencing of DNA at different time points gives rise to a frequency distribution, where the modes give you the Origin points.
  • This method can also give you the speed of the fork , etc.
• With Cloning we can find all potential Origin sites, however, with Deep Sequencing, you find the Origins that are being actually used by the cell.
• End replication problem: Each time a cell divides, the chromosomes should be shortened. Finally the chromosomes should and be damaged.
• The telomere: Its a end point sequence that protects linear DNA from degradation by replacing the shortened DNA bases.
• Telomerase: the enzyme that does the replacement.
• Some cancers can express Telomerase to bypass the cell division limit.
• Telomers are repeated sequences. The T-loop hypothesis states that the end of the DNA strand loops back to itself to make a loop, so as to not signal as a broken DNA strand. The end-looping region is the Trlomer region.
• Different organisms also use different mechanisms to deal with end poins of its genetic material.

Day 5

Centromeres: from epigenetic to genetic

• Cell division has been studied for a long time because of its importance in lifes’ processes.

• Centrosomes and centromere are the two key factors for cell division.

• The centromere is a protein/DNA structure which links chromosomes to spindle microtubules.

• Centromeric architechtures are diverse:

  • Regional centromeres
  • Point centromeres
  • Holocentromers

• Kineticore: Protein complex TODO

• If there are problems with Kineticores, you get aneuploid. ~90% of solid human tumors and ~75% of hematopoietic cancers are aneuploid.

• Key steps for a centromere:

  1. Centromere Position (decision on the location of the centromere, which is permanent)

  2. Centromere function

  3. Centromere integrity

• The 171-bp alpha satellite is the fundamental unit of the centromere DNA.

• ~3% of the total human genome is chromomeric DNA. CENP-B is the only centromeric DNA-seq specific binding protein.

• The centromeres exist in two domains: CENP-A and CENP-B.

• Bigger centromere regions can lead to tangling of microtubles and chromatin during division.

• Histones:

  

• CENP-A and CENP-B arer particular histones that are specific to centromeres.

Where do Chromosomes go in interphase?

• Interphase: Period between the cell divisions. During this phase, the genetic material (DNA) is replicated by the cell.

• In mammals, you can find chromosomes that like to localize in a certain region of the nucleus.

How to build a mitotic chromosome?

• mitosis means “threads”. The threads are the chromosomes.

• Why cells need to build a mitotic chromosome?

  • spatial compaction
  • mechanical properties
  • resolution of sister-chromatids
  • switch off transcription

• Interphase to mitosis compaction: 2X compaction.

• Assembly of mitotic chromosomes is not fully known. There are several models:

  • Hierarchical folding
  • Scaffold/radial-loop
  • Chromatin network
  • Thin Layer stack

• The localization of chromatin position explains how features far away in the genetic sequence regulate each other.

• Positive supercoils are formed when the Replication Fork travels through the chromatin strands. To release these tension:

  • Rotation of the replication fork itself.
  • Nick the strand to release the tension and then re-attach.
  • Topoisomerase II is required for decatenation and sister chromatin resolution

• Transcription is turned off during Mitosis. Dogmas:

  • Chromatin accessibility is limited.
  • Global transcription suppression.