is SMART only suitable for pure compounds?

[lfjl...@gmail.com]

Dear Sir/Madam,

my name is Fengjie Li and I am a postdoctoral researcher in NPs in GEOMAR. I was so excited when I saw this new technique. One question that I would like to ask is: will SMART only be suitable for purified compounds? When I read the article "Small Molecule Accurate Recognition Technology (SMART) to Enhance Natural Products Research" published on Scientific Reports, I have the feeling that this technique only applies to pure compounds. But when I saw the more recently published article "A Convolutional Neural Network-Based Approach for the Rapid Annotation of Molecularly Diverse Natural Products" published on JACS, SMART was applied to the most cytotoxic crude fraction H. So here I am very confused. In a mixture, how SMART gonna differentiate HSQC features from different chemicals?

 

Hope to discuss with you in detail if possible.

 

Best,

Fengjie Li

[beowu...@gmail.com]

Hi Fengjie,

Thank you very much for your interest in SMART. SMART correlates an HSQC spectra pattern with its nearest analogues in our training dataset using deep learning techniques. That is how the tool works. By testing pure compounds with SMART, we found out that SMART is able to associate an unknown compound never used for training with analogues that contains similar substructure moieties. I.E., compounds with thiazole rings. Naturally, we consider that SMART can be used to detect compounds in an HSQC of a mixture. Deep convoluational neural networks can be used for image segmentation. Specifically, it can be trained to identify names of each person in a party scene of a meeting scene. But people are still trying to understand what features the algorithm extracts.

Best,

Chen

[Raphael Reher]

Dear Fengjie Li,

Here some additional information to Chen's answer. Our new version SMART 2.0 that was published in JACS has a much larger dataset (53,076 vs 2,048 natural products) and also the training of the CNN was done a little differently.

The main application for SMART 2 is fast dereplication of known natural products and assistance with the structure elucidation of new natural products.

However, I just gave it a try to feed mixtures to SMART 2.0 and I got useful results also for other projects, but not in every case. Here is my thoughts and practical tips regarding mixture analysis with SMART 2.0:

- mixture analysis is not the main purpose of SMART 2.0 yet, so everything I tell you is preliminary. We are working hard on automating and facilitating mixture analysis though

- SMART seems to be fairly robust towards different NMR solvents as well as missing and additional CH-correlations of compounds, which is a prerequisite for its use for mixture analysis

- in mixtures you can only expect from SMART to detect the main metabolite (I would interpret the results as getting hints to the major structural class of NPs in that fraction). In my case fraction H had a major metabolite, in cases of crude plant extracts for example it might be more difficult.

In general the bias regarding mixtures is, the more CH-correltations the more suggestions for high-molecular weight compounds you will get. However, I even got reasonable SMART 2.0 results for a crude bacterial extracts with 297 CH correlations. Just do not expect that you will get the correct structure out of

crudes, but more like interesting structural elements.

- you might have to adjust aquisittion times of the HSQC, depending on the weight of your fraction

- you should create many different NMR tables for the same sample (adjust the gain to see only the main metabolite, and then compare results when you increase the gain (you will increase signal and noise and get much more correlations).

I would say, give it a try and please report back if you have any troubles and also if it worked well for you!

Thanks & Best regards,

Raphael

On Monday, March 9, 2020 at 7:32:12 AM UTC-7, lfjl...@gmail.com wrote:

[lfjl...@gmail.com]

Hi Chen,

thanks for your replay.

I will try it out and let you know.

Best,

Fengjie

[lfjl...@gmail.com]

Hi Raphael,

thanks a lot for the detailed explanation. I will try it out with my crude extracts or/and fractions. I will come back to you once I had any updates.

Best,

Fengjie