1. Relating Small Molecule Structure to Small Molecule Performance
Establishing structure-activity relationship (SAR) for novel disaccharides exposed to preadipocyte cell states
Hi, my name is Isaac Joseph and I’m a pre-freshman at MIT. This summer, my project has been relating small molecule structure to small molecule performance. This is called a structure activity relationship. As you saw in the previous presentation, finding the correct small molecule to produce a particular effect is useful in many situations. The data that I used to develop my methods were disaccharide small molecules used in preadipocyte (pre-fat-cell assays).
Isaac Joseph

2. Why is finding SAR useful?
Drug development
Effect enhancement through structure optimization
So why should we find SAR’s? Two major applications are in the pharmaceutical industry and in chemical biology research. Each activity involves the action of small molecules on living systems. During development of drugs or research tools, we often use SARs to learn how to modify small molecules to optimize their effects.

3. How did we previously assess SAR?
Visual inspection of data and structures
Introduces bias
Time -consuming
Computational analysis – often used with isolated proteins
Doesn’t account for cellular context
Usually deals with only one biological outcome
How is this currently done? It’s mostly done by a bunch of chemists getting together and basically saying ok: we think X part of the molecule is relevant to Y effect, so maybe we should change X and hopefully Y will change. One problem with this is that it introduces bias; by science’s nature, the best discoveries are not always the expected ones; also, it’s time consuming because every single molecule has to be examined individually. There is some computational analysis done, but this is only done on an isolated protein level. Once the proteins are put into cells, they behave differently; also, since it’s only one protein, only one outcome can be assessed at once.

4. How do I assess SAR? Describe structure and performance
Alter structure description to match performance
Interpret results
So what’s my method of finding SAR that accounts for all these problems? First, for an assay, I describe the structure of the molecules and the effect seen with these molecules computationally. Then, I simply alter the structural description until it most highly accords with the performance seen with the same molecules. Once structures are chosen, we then interpret the results for optimizing the molecules.

5. How do we describe elements of small molecule structure?
regiochemistry
structural descriptors
How do I represent structure chemically? Since the molecules all share certain features, I simply represent each difference as a descriptor. There are three kinds of diversity in the library we studied; here’s regiochemistry.

6. How do we describe elements of small molecule structure?
stereochemistry
structural descriptors
Here’s stereochemistry.

7. How do we describe elements of small molecule structure?
appendages
structural descriptors
And here’s appendage diversity.

8. How do we represent small molecule structure computationally?
structure matrix
structural descriptors
64 small molecules
Now, I take these descriptors and represent them in a binary structure matrix. Each row is a structural fingerprint of a specific compound, and the red cells indicate the presence of a certain structural feature.
50 descriptors
presence
absence

9. What biological effects were measured?
5 preadipocyte cell states
Wild-type brown preadipocyte
IRS 1 -/- brown preadipocyte
IRS 2-/- brown preadipocyte
3T3- L1 white preadipocyte
Tunicamycin modified white preadipocyte
2 measured effects
Lipid accumulation
Mitochondrial membrane potential
What effects were measured? We took 5 states of preadipocytes – cells that will differentiate into adipocytes, which are fat cells – and measured two effects – differentiation and mitochondrial membrane potential.

10. How do we represent biological effects computationally?
performance matrix
64 small molecules
We represented these measurements in a performance matrix analogous to the earlier-shown structure matrix. Here, each row is the fingerprint of a compound in effect, where a red cell represents a particular effect.
80 assays (2 columns each for +, - effects)
effect
no effect

11. How do I assess SAR? Describe structure and performance
Alter structure description to match performance
Interpret results
So now we’ve described structure and performance computationally. To figure out what parts of structure are relevant to performance, we simply alter our structural description until structure similarity of compounds aligns with their biological performance similarity

12. Step 1: Calculate similarities between small molecules in performance
performance matrix
64 small molecules
pairwise performance
similarity
80 assays
To do this, we first create a similarity matrix out of the performance matrix, to find the similarity of between each compound in this realm. Each cell in the performance matrix represents the similarity value between two compounds.
effect
no effect
64 small molecules
64 small molecules
dissimilar
similar

13. Step 2: Calculate similarities between small molecules in structure using partial description
8 descriptors
64 small molecules
candidate structure matrix
structure matrix
64 small molecules
50 descriptors
pick a subset
64 small molecules
pairwise structure similarity
Then, we choose random structures to consider, and create an analogous structure similarity matrix of each compound.
presence
absence
dissimilar
similar

14. Step 3: Check similarity between performance and structure
pairwise performance similarity
64 small molecules
pairwise structure similarity
VS.
64 small molecules
We then see how similar compound performance is to our compound structural representation.
64 small molecules
dissimilar
similar
dissimilar
similar

15. Step 4: Change structural description until similarity between performance and structure is maximum
pairwise performance similarity
64 small molecules
pairwise structure similarity
optimized pairwise structure similarity
We repeatedly change our structural representation until we find one that best accords with performance. Then, find out which selected descriptors that gave this similarity; these have the most to do without our performance in question.
all descriptors
selected descriptors

16. How do I assess SAR? Describe structure and performance
Alter structure description to match performance
Interpret results
So now that we’ve found the structures that have the most to do with performance, what can we do with this data?

17. How do we know it’s working?
performance matrix
structure-performance similarity (correlation coefficient)
columns permuted
22 times
p < 0.046
First, how can we be sure that our results are significant? be sure that our results are significant?  To test this, we randomly permuted the results from each assay and selected a structure description, then compared the results from many such permutations to our best description using the real performance data and returned a p-value of <0.05, which means that there’s less than a 5% chance that our results were obtained by accident.

18. What do the selected structures mean?
selected descriptors
optimized structure
similarity matrix
performance
similarity matrix
With the chosen descriptors, we can then create structural drawings by looking at similar blocks of small molecules chosen from the similarity matrix created from the best structural elements. If we want to optimize the effects of these similar small molecules, we should keep the chosen structures as constant (because these are already creating the desired effect), and modify the rest to see if the effect can be increased further. Here I show the disaccharides from my particular data, but we hope to expand our method for optimizations of small molecules for all purposes.
8 small molecules

19. Acknowledgements Analysis Paul A. Clemons (mentor) Joshua C. Gilbert
Nicole E. Bodycombe
Chemistry
Micha Fridman
Tetsuya Tanikawa
Daniel Kahne
Biology
Bridget K. Wagner
Administration
Shawna L. Young
Bruce Birren
I’d like to thank the following people who were vital to my project…. The people that helped with analysis, especially my mentor Paul Clemons, the biologists, administration, and the people that actually created the molecules, the Kahne group at Harvard’s Department of Chemistry & Chemical Biology.