Thread by @WeAreJADBio

Going to break-down how easy it is to use #autoML and more specifically JADBio AutoML. If you need an account to try it out, head over to jadbio.com and grab a free Basic plan. Ready? #data #ML (1/16)

STEP 1: You start by creating a Project on JADBio and generating all your study #data. That could either be data that has been processed and normalized by a #Bioinformatician, or public data available in the known data repositories (2/16)

If you’re using software for #molecular diagnosis like our Partner’s @QIAGEN OmicSoft Lands platform, your data is ready to be uploaded on JADBio. #ML 3/16

Then again don’t forget you can use your own research data, publicly available data or utilize the datasets available on JADBio. #AutoML #bioinformatics (4/16)

STEP 2: Perform your Predictive Analysis. Select the outcome to predict by. Choose between #Classification, #Regression and Survival Analysis. #ML #survivalanalysis > We'll talk about that in future tweets. (5/16)

#Classification if the outcome to predict is binary e.g. case-control, categorical or different cancer types. #Regression if the outcome is continuous e.g. the viral load. Or #SurvivalAnalysis if the outcome is the time to an event e.g. death, relapse, etc. #ML (6/16)

Ok up to here? Let’s continue. Then select what predictive performance metric would you want to optimize for? AUC, Accuracy, F1 Score, R2, Mean Squared Error, Concordance Index (7/16) #ML #metrics #machinelearning #autoML

Basically, that’s it! Your ML model is in the making. Obviously, that’s not enough & you now have to do something with it! I promise it’s highly credible & its application on your new data will verify it! Grab a free account to do all this yourself jadbio.com 8/16

Here’s where all the unleashed power of JADBio comes in!
Discover Knowledge> 1. Automatically get the biomarkers that collectively lead to optimal predictions, and 2. Examine their role in predictions and added-value within analysis results #biomed #clinicaldata 9/16

The analysis of your data produces a wealth of visuals, plots, and graphs for interpretation of the model. I know, I know... I’m showing off, but seriously proud of our wealth of plots, graphs, reports. We can fill a whole research paper with visualisations! (10/16) #AutoML

Starting w/
a. ROC and Precision-Recall curves for optimizing classification thresholds
b. PCA / UMAP plots on the signature markers showcasing sample distribution as per the predictive model 11/16 #ROC #PCA #visuals

and
c. Predicted Probabilities Distribution plot for each class (categorical outcomes)
d. ICE graphs visualizing the prediction dependency on specific feature values
e. Kaplan-Meierr curve for stratifying to low/medium/high risk groups, and many other graphs/plots 12/16 #plots

STEP 5: (Yes, I know I jumped 3 & 4… but no I didn't just forgot to add the steps & Twitter doesn’t allow corrections)
So now you have your #ML model ready! The question is How do you intend to apply the trained model? 13/16 #machinelearning

There are several things you can do with it!
1. Validate model against labeled samples
2. Predict outcomes for unlabeled samples 14/16
#machinelearning #autoML #model #predictions

and
3. Explore what-if scenarios (manually entered predictor values)
4. Download the model to make predictions offline. Check out what these guys from the SU Dept. of Bioinformatics did with their JADBio model covidoutcome.com 15/16 #ML in practice!

Want to do it yourself and get some insights from all the data you have collected? Seriously, you can make the next big treatment or drug discovery, that will save some part of humanity. Even win a Nobel Prize for Medicine ;) Go for it! Jadbio.com 16/16 #AutoML