Miscellaneous

Question 1

In humans, how much does the size of a gene vary?

Answer 1

From a few hundred DNA bases to more than 2 million

Question 2

How many chromosomes, base pairs and genes in human genome?

Answer 2

23 pairs chromosomes
3.2 billion base pairs (6.4 billion bases)
25,000 genes

Question 3

How many genes in human genome?

Answer 3

20,000-25,000

Question 4

Difference between intron and exon?

Answer 4

Introns are non-coding DNA sequences within a gene that are removed by RNA splicing during maturation of the RNA product.

Exons are protein-coding DNA sequences that have the necessary codons for protein synthesis.

Question 5

How big is a typical IgG mAb (in kDa and number of aa)?

Answer 5

Roughly 150 kDa and 1400 amino acids long

Question 6

How is a mAb’s binding surface constructed?

Answer 6

The mAb’s binding surface is the paratope, made up of 6 CDRs or distinct variable loops: 3 on the light chain (L1, L2, L3) and 3 on the heavy chain (H1, H2, H3)

Question 7

Why is the H3 loop of the CDR so challenging?

Answer 7

H3 loop is challenging because it has great diversity in length, sequence and conformation

Question 8

What was the first tumor-agnostic FDA approval?

Answer 8

In 2017, FDA approved pembrolizumab (PD-1, Keytruda, Merck) for patients with any metastatic solid tumor who are microsatellite instability-high (MSI-H) or mismatch repair deficient (dMMR)

Tumors that have a dMMR system can develop MSI. Microsatellites are regions of repeated DNA that change in length (show instability) when mismatch repair is not working properly

Question 9

What are two main types of immunogenicity?

Answer 9

1) Driven by T cells (easy to identify by AI)
2) Driven by immune complex formation (typically because of aggregation)

Question 10

How many known protein, protein 3D structures, protein-protein complexes vs antibody-antigen complexes are there in the PBD?

Answer 10

200 million known proteins

≈200,000 protein 3D structures
≈100,000 protein-protein complexes
3000 antibody-antigen complexes

Question 11

How is TCR (octameric) complex organized?

What are CD4/CD8?

Answer 11

2 TCR receptor chains (alpha and beta in 95% T cells; gamma and delta in 5%) that form the ligand-binding site (diversity); associated in membrane with CD3 (contains 4 distinct chains; 6 chains in total: epsilon, delta, gamma, epsilon) and 2 zeta chains

CD4/CD8 are co-receptors

Question 12

What are 3 signals for T cell activation?

Answer 12

1. Antigen recognition (peptide:MHC binds to TCR/CD4 (or CD8); typically takes place in secondary lymph organs)
2. Th cells: co-stimulation (CD28) binds to CD80 (B7.1) or CD86 (B7.2) on APC –> T cell proliferation. Cytotoxic T cells: co-stimulation (CD70 and 41BB/CD137)
   Lack of signal 2 leads to anergy/tolerance
3. Cytokines

Question 13

First TCE to be FDA-approved?

Answer 13

Blinatumomab (Blincyto), CD3/CD19, approved in 2014 for R/R B-ALL. Short half-life (2 hrs) and requires continuous IV Infusion

Question 14

How many codons exist?

Answer 14

There are 64 different codons: 61 specify amino acids and 3 are used as stop signals.

Question 15

How many GPCRs are there, and how many are drug target opportunities?

Answer 15

Approx 800, over half are sensory/olfactory, with the remaining 370 presenting drug targeting opportunities

GPCRs as a drug class represent 34% of FDA approvals

Question 16

What are the two types of epitopes?

Answer 16

Continuous residues –> linear epitope
Discontinuous residues –> conformational epitope

Question 17

Describe the pMHC complex

Answer 17

pMHC complex consists of: HLA class I heavy chain, beta2-microglobulin and peptide

The MHC class I heavy chain consists of three extracellular domains (α1, α2, α3), a transmembrane domain, and a cytoplasmic domain. β2 microglobulin forms a fourth extracellular domain and is held in the complex by noncovalent interactions.

Question 18

How many aa exist and what are the major classes?

Answer 18

20 natural amino acids
Major classes: hydrophobic, aromatic, charged, polar

Question 19

How is 10^36 search space calculated?

Answer 19

20 natural aa over ~30 position of the H1/H2/H3 loops to give 20^30 ~10^39 combinations. Then minus some combinations since those clearly don’t make valid antibodies.

Question 20

What is the repertoire of unique sequences of antibodies in the human immune system?

Answer 20

10^13

Question 21

What is the RMSD in structural biology?

Answer 21

The root-mean-square deviation (RMSD) is a measure of how accurate a prediction or model is compared to a reference or target structure. A lower RMSD value indicates a higher level of accuracy in the prediction.

Question 22

What is the Rosetta classical scoring function?

Answer 22

Calculation of the energy/score of a protein structure based on its 3D shape and physical and chemical properties (bond angles, distances etc).

A lower score is better (indicates more stable and energetically favourable structure)

Question 23

What is a diffusion model?

Answer 23

Diffusion models are generative models that work by adding noise (Gaussian noise) to the available training data (aka the forward diffusion process) and then reversing the process (aka denoising or the reverse diffusion process) to recover the data

Question 24

How is a latent space a lower-dimensional representation of the antibody structure?

Answer 24

In the case of antibody structures, the latent space is a mathematical representation of the structure that captures the most important features of the antibody (e.g. overall shape, positions of the amino acids and interactions between the amino acids).

Question 25

How are camelid and shark antibodies unique?

Answer 25

Camelid: no LC, only VH domain (and no CH1 domain); CDR-H3 loop longer than human mAb H3

Shark: VNAR (variable domain of new antigen receptor) are smallest ag-binding domains found in nature; no LC
Each H chain: 5 CNAR domains + 1 VNAR. HC does not have a CDR-H2 (only H1 and H3)

Question 26

What is a generative model?

Answer 26

Generative modeling is a way for computers to learn how to create new things that are similar to things they’ve seen before (i.e. the training data)

Question 27

What are:
- ProteinMPNN
- AbFormer
- AbFold

Answer 27

ProteinMPNN: external benchmark, published and used out of the box (Baker lab)

AbFormer: LLM (sequence only, no structural information)
Generate CDR sequences that are similar to hu CDR sequences
Masked-language model; trained by randomly masking an aa and learn how to reconstruct missing aa in that sequence; trained on 1.3 billion VH sequences from 80 studies (OAS) –> 36 million unique CDR3 sequences + 1 million unique CDR1 + 1 million unique CDR2 sequences.
Then fine-tune AbFormer on therapeutic antibodies

AbFold: H3 design tool based on AlphaFold and AbFormer

Question 28

What are the Fable Foundation Models (Fable-FM)?

Answer 28

Fable-Abformer (enhances manufacturability and ensures properly folded abs)

Fable-RE:
1) Equivariant transformer (global frame vector graph-based transformer)
2) Transfer learning (pre-training on large sets of protein structure data; to overcome data scarcity challenge)

Question 29

What are generative and semi-generative models built on top of Fable-FM?

Answer 29

1) RE-Dock = antibody pose on epitope; RE-Diff-NG = will generate poses and loops (Q3)

2) RE-Diff/Abfold = CDR loop conformations (fully de novo CDR loops)

3) RE-Masked lang model = sequence / side chain conformations

Question 30

Fable pitch deck: what does ab design require?

Answer 30

1) Antibody structure compatible with high affinity binding
2) Antibody sequence that is developable

Question 31

Fable pitch deck: what are four thing going on at same time?

Answer 31

• Ab pose on epitope (identify ab pose on chosen epitope suitable for high affinity binding)
• CDR loop conformations (given ab pose, design CDR loop conformations for high affinity binding)
• Sequence/side chain conformations (given pose and CDR loops, identify suitable side-chain sequences)
• Manufacturability (choose sequences suitable for human antibody-like behavior)

Question 32

Why is it easier to in silico predict T cell epitopes compared to B cell epitopes?

Answer 32

T cell epitopes are linear peptides presented in context of MHC, B cell epitopes can be conformational (i.e. structural folds in addition to specific aa sequences)

Question 33

How many total T cells are there in a human, and how many unique TCRs are expressed within this population?

Answer 33

~3x10^11 total T cells
~1x10^8 unique TCRs within this population

Question 34

What is self-attention mechanism?

Answer 34

Allows LLMs to weigh importance of different words in a sentence when generating text

Question 35

What is Werewolf Tx’s platform?

Answer 35

Conditionally activated cytokines (Indukines): systemically delivered as inactive pro-drugs. Indukines are a single molecule containing 1) an inactivation domain, 2) a half-life extension domain, tethered to 3) a fully active IL-2 (tethered by protease-cleavable linker) ​

IL-2: Ph 1 (onc)​
IL-12: Ph 1 (onc)​
IFNa: Ph 1 (onc, Jazz Tx)​
IL-21: IND (onc)​
IL-18: discovery (onc)

Question 36

What are Bonum Tx’s two ATP-dependent programs?

Answer 36

ATP-dependent cytokines are inactive in circulation; become active with ATP present in TME​

ATP-IFNa​

ATP-IL12​

Question 37

What are Bonum Tx’s conditionally activated biologics programs?

Answer 37

PD-1/IL-2 (Roche)​
PDL1/IFNa​
LAG3/IL-2​
LRRC15/IFN-a (LRRC15 targets CAFs)​
LLRC15/TGFbR2

Question 38

What are Biolojic’s four programs?

Answer 38

AU-007: mAb specific to the CD25 (IL-2Ra) binding site on IL-2 – prevents IL-2 binding to trimeric receptor (Treg), while allowing binding to dimer receptor (Teff, NK) (Ph 1, Aulos spinout)​

IL-13/TSLP dual-specific ab (IND, Teva)​

Treg agonist (IL-2 mAb specific for CD122/CD132 epitope, leaving IL-2 free to bind CD25 on Treg) (discovery)​

TNFR2 agonist (with Nektar)

Question 39

What are Bonum Tx’s ATP-dependent programs?

Answer 39

ATP-IL-12
ATP-IFN-a

Question 40

What’s in Synthekine’s pipeline for I&I?

Answer 40

Ph 1: IL-2 + CD19 CAR (no lymphodepletion)
Preclinical: engineered IL-10 (partial agonist)
Preclinical: engineered IL-22 (partial agonist)

Question 41

Where do mAbs bind to FcRn?

Answer 41

IgG monoclonal antibodies (mAbs) bind to the neonatal Fc receptor (FcRn) primarily in the acidic environment of endosomes within cells. This interaction occurs mainly in the endothelial cells of blood vessels and in epithelial cells of tissues such as the intestine and kidney.