Using Computer Code to Decipher Genetic Code - Part 2 (Bioinformatics 101)

Key Takeaways

welcome back to the data professor YouTube channel if you new here my name is Shannon not a sin Ahmad and I'm an associate professor of bioinformatics on this YouTube channel we cover about data science concepts and practical tutorials so if you're into this type of content please consider subscribing okay so let's compare between bioinformatics and data science and so to the left by informatics is aimed at making sense and gaining insights from biological data and so data science is just more generic term to mean make sense and gain insights from data and so we can see that bioinformatics and data science are quite similar in the sense that biology is replacing domain knowledge and so statistics and computer science here are essentially the same and so the domain knowledge in bioinformatics it's essentially biology ok so this is just a crude comparison between both field and bio formatic might not entail the use of machine learning algorithms and there are other parts of bio formatic that are not accounted for in this simplified comparison just a point to note ok so you might ask why do we need computational models for drug discovery so let's look at some case study of other areas so computers such as IBM D blue had defeated human beings in jeopardy and chess Google released a self-driving car nASA uses computers to simulate space mission computers are being used to design aircrafts and cars supermarkets and shopping malls are using purchase history to analyze and predict our spending behavior for example the membership card that they gave us they could track our purchase and then analyze the data for example buy Market Basket analysis by seeing which products do we normally buy together and also Amazon go has made it possible to use AI and face recognition to allow shoppers to just walk in walk out right just walk in grab what you need put it in your bag and this leave and so with all of these case study in mind the natural question would be why not use it for discovering this and developing new drugs and so the thing is we do we do use computers for drug discovery and so the first example is to discern the structure-activity relationship of chemical library and so this means that the chemical structure of compounds or molecules are being correlated to their molecular features which is a quantitative or qualitative description of the chemical structure and so such computational model such predictive models are now being used as alternative to some experimental work and a notable example here is the use of deep learning to decode and encode Smiles notation of molecules in order to analyze the structure activity relationship and based on such model you said to generate new molecules so this is a very interesting application and on the right here we see that there is the PCA analysis or the principal component analysis which tells the clustering of the active drugs and the inactive drugs meaning compounds that have good activity toward the target protein and bad activity toward the target protein so green and red so Green is good red is bad okay so additional examples computational models can be quickly built to predict the pharmacokinetics and bioactivity of query compounds by pharmacokinetic it means the absorption distribution metabolism excretion and toxicity of drugs and the bio activity is the inhibition or activation of the target protein and so the data that you have collected in part one essentially falls into this bullet point particularly the bio activity of query compounds okay so the compounds that you have compiled which inhibit the corona virus are compounds that have been tested experimentally to inhibit the protein from the corona virus and so you could change the name of this target protein to other protein such as aroma taste or other protein that you are interested in and so the compounds are modulating and by modulating I mean they are trying to control the target protein by either inhibiting it or to activate it okay and so you could think of these molecule as the on and off switch of the target protein so when you apply that molecule you could turn it off or turn it on in terms of the function of the protein and so your predictive model using machine learning will allow biologists and chemists to understand the relationship between which molecular features give rise to the biological activity and such predictive models can be applied for personalized medicine and so this figure is an example of the functional group analysis from structure activity relationship model building from one of our papers and other specific questions that can be answered by computational models include what target proteins what target proteins can my compound bind to and so the target protein could be the aromatase or it could be the protease it could be the glucose CDs amylase etc and so with the target protein of your interest you want to know which compound or small molecule could come and modulate this target proteins activity either to inhibit or to activate it and so the second question could be what type of compound can bind and modulate the bio activity of the target protein of my interest and the third question would be are there any similar compounds to my query compound that may potentially exert similar binding behavior so let's say that you have a fda-approved drug that has been known to inhibit aromatase enzyme and you want to know is there any other small molecule that have similar structure with this fda-approved drug and so the reason being that the fda-approved drug might be a good drug but it might have some side effects that are undesirable and so the goal of drug discovery and drug design is to develop drugs with minimal side effects and so that entails the optimization of the pharmacokinetics comprising of the absorption distribution metabolism excretion toxicity okay and so it would be interesting to see is there any similar structure to the fda-approved drug that have similar binding effects but with a safer pharmacokinetic profile and so all of these could be answered by applying machine learning and data science okay and so as we know data science is the process of identifying and making sense of hidden patterns and so this hidden patterns could be the knowledge and we try to make sense of hidden patterns that are found in the large amounts of data and so typically the data has this hierarchy going from raw unstructured data to becoming more structured data and then once we structure the data we could uncover the patterns and once we have the patterns we can gain knowledge and finally we could apply it to have wisdom okay and so on to QSAR modeling which stands for quantitative structure-activity relationship modeling so this is a mathematical modeling that tries to find relationship between chemical structure and the bioactivity or the biological activity and so the chemical structure can be represented by a set of molecular descriptors which could be quantitative or qualitative and the molecular descriptor will be about the physical chemical properties of the molecule the molecular descriptor could either be global features or local features by global meaning molecular weights at the holistic level and local features could mean like the small minor detail like the functional groups hydroxy group nitrogen atoms or the charge a specific portion of the molecule okay so I'm gonna cover that more in the subsequent video tutorials where we generate them a little descriptor and so I'm gonna tell you which one are the global features and which ones are the local features and so this is the workflow of the QSAR modeling or accuser modeling and so this is from the first review article that I have written back in 2009 so it was my first review article which I wrote about the Q star modeling so you could copy this and paste it into Google to read more details about it or I could also provide it in the description down below and I could also provide it in the video description so you can check that out and another similar terminology related to the Q star modeling would be pro do chemometric modeling and so I'm not gonna go into much detail about this it's more advanced which would be better to save for future videos as well and so let's think of the pro you chemometric model as several QSAR models combined together okay so that's the essential concept of the PCM so we're gonna skip it for now and so this is the holistic level of all of the resources and tools available for drug discovery at the holistic level and so this is from one of my editorial articles and this is a summary of the procedures for the development of the QSAR model so it's essentially the development of machine learning models and when we use machine learning to make sense of this chemical data biological data we change the name to QSAR but it's essentially machine learning model so as you can see we have data collection biological data collection we generate the descriptor feature generation feature selection data pre-processing splitting of the data to training and test set validating the model internally and developed a model and performing evaluation of the model performance and also to perform external validation on an external sets okay and so this is a list of the chemical databases and the list of molecular descriptor softwares and so for future tutorial videos of this bioinformatics project series we're gonna use free software or open source software so don't worry that you have to buy expensive software for following along so we're gonna make use of the open source software and this is the list of the computational chemistry software so don't worry about that so back in 2016 we organized the first international conference on pharmaceutical bioinformatics so it was a collaboration with Uppsala University so we bring together into ciass of drug discovery and design and so this was the poster advertisement for the conference and yours truly and so the question here is why do we need to develop our own bioinformatics tool so you might notice that there might be several thousands of bioinformatics tools that are already in existence so do we think that all possible tools would have already been developed what do you think is it true or false so let me know in the description and we could discuss about this and the second question is bioinformatics tools will be available forever will it be available forever let me know in the comments true or false existing tools may lack certain features that we need in our own project what do we do do we develop our own tools or do we proceed without this feature so just go ahead and ignore the feature that we wanted okay so some more questions for you which path will you take will you hire a programmer to develop the bioinformatics tool or will you learn how to program ok so there's two possible answers so let me know in the comments down below which is your answer so these are selected web servers and software that we have developed in our research group and so these software as you can see are all related to by informatics and the first one was wrapping the weaker software inside Python in order to automate the development of neural network models and support vector machine models so actually we built this back in early 2010 or 11 and it was the time when no automatic ml was available and so perhaps it was among one of the first automated data mining software and the software here was mentioned in one of the book chapters published by Springer in the book artificial neural network and we also develop other bioinformatics web server including OSF P cryoprotectant pret bio curator and Pi max and so our research group are practicing research reproducibility whereby we try to share the code and data that were used to prepare the analysis in the papers that we published so that other interested users can reproduce the work and perhaps make use of it in their own research and these are some of the book chapters and review articles that we have contributed and so this is the bonus I was talking about earlier on which is the steps to developing a bioinformatics tool and so as you can see essentially there is six steps and so the first step would be to come up with the concept of the bioinformatics tool so you want to figure out what bioinformatics tool or software that you want to develop and normally the idea for the bioinformatics tool will come from an unsolved problem that you might have or it might be an inconvenience to your project that you figure out would be an interesting topic to explore further for example it could be a problem or something that might slow down your analysis and if you are able to solve that by means of developing a bi automatic tool that it could not only save your time but other people's time as well okay so coming up with the concept of the bottleneck tool could come from the problems that you encounter and so once you come up with the tool you want to make a list of the features that you want to see so make a wish list of the features and so step number three is to list the sequential workflow methodology meaning the protocol or the pseudocode of the bioinformatic tool like step one step two step three what do you do how do you process the data okay and so step number four is that you have to realize that bioinformatics tool are essentially a collection of small tasks meaning that if you click on one button it will invoke a particular task it could be performed by a particular function that you develop and if you click on another button it will perform another function right so it is essentially a collection of tasks and so you weave it together to get this software so step number five is you want to work on the coding of each of these small tasks individually and so let's think of it as like a chapter in a book right so many chapters will comprise a book so you will work on individual chapters and within no time you will be able to complete all of the chapter to form the entire book and so the last step is to make sure that the entire workflow were as desired okay so this is about testing about debugging so you want to try from the beginning you want to input the data and see whether it finds the desirable intermediate data and the output data which is the final data right okay and so if you find value in this video please give it a thumbs up and if you haven't yet subscribed please subscribe to the channel and as always the best way to learn data science is to do data science and please enjoy the journey thank you for watching please like subscribe and share and I'll see you in the next one but in the meantime please check out these videos