On-Demand Session: New Machine-learning-driven Opportunities for AFM Data Acquisition and Analysis

       _{PRESENTATION HIGHLIGHTS:}

• [00:00:40] Using AFM to measure cell shape as biomarkers of biochemical response
• [00:04:11] Advantages of using AI-guided AFM for live cell research
• [00:04:55] Generating phase-contrast images
• [00:06:22] Detecting cell shape
• [00:09:41] Training the neural network with transfer learning]
• [00:10:45] Example cell shape detection results
• [00:13:46] Navigating the cantilever tip to the location of the shape
• [00:18:10] Performing EFM imaging
• [00:23:30] Conclusions

_{PRESENTATION HIGHLIGHTS:}

[00:00:40] Part 1: Using machine learning to classify and predict structural property relationships from AFM images

• [00:04:20] Data collection and pre-processing
• [00:09:17] Classify AFM images of ICPs
• [00:10:59] Compare phase data & PFQNM Data

• [00:12:10] Structure property predictions
• [00:15:55] Part 1 summary

[00:17:21] Part 2: Artifact detection research

• [00:17:53] Experimental overview
• [00:19:56] Machine learning approach
• [00:21:07] Structural similarity approach
• [00:23:39] Part 2 summary

       _{PRESENTATION HIGHLIGHTS:}

• [00:00:11] Sub-molecular imaging of DNA
• [00:00:44] Development & capabilities of novel AFM probes for imaging the DNA double-helix
• [00:05:05] How can we quantify information from AFM?
• [00:09:36] Introduction to TopoStats: automated quantitative AFM analysis
• [00:14:09] Training TopoStats for unsupervised learning for molecule identification
• [00:17:23] Using TopoStats to study the effect of spercoiling on DNA structure
• [00:20:34] Using TopoStats for quantitative molecular analysis
• [00:24:27] Results of using TopoStats to study the effects of supercoiling on DNA structure
• [00:26:39] The future of TopoStats: Base-pair resolution analysis of the effect of supercoiling on DNA structure and flexibility

       _{PRESENTATION HIGHLIGHTS:}

• [00:00:16] Why and how to use machine learning with AFM

Collect and process data for machine learning with Bruker technology

• [00:05:06] AutoMET Automation Software
  
• [00:06:50] User-friendly API in the NanoScope
• [00:10:07] What is "good data" for machine learning with AFM?
• [00:11:07] Collecting "good data" with Tapping Mode and PeakForce Tapping
• [00:12:23] Example: Using AutoMET to investigate mechanical properties of polymers
• [00:15:03] How to import raw nanoscope data into Python
• [00:17:59] Summary

_{PRESENTATION HIGHLIGHTS:}

Presentation 1:

• [00:00:00] How were the cells in your phase contrast images labeled?

Presentation 2:

• [00:04:16] Was the image curation process Human-based or automated?
• [00:04:51] How big should the dataset be in order to build a machine-learning model
• [00:07:05] Why does the deformation channel perform best in the CNN? How to gauge how many training sets are needed for reliable CNN performance?

Presentation 3 

• [00:09:54] What parameters were used for clustering analysis?
• [00:11:09] Does TopoStats work with raw Bruker data files?

Presentation 4

• [00:12:00] In the nanoscope software, and using peakForce QNM mode, how can you get representative force-distance curves over a scan area?
• [00:13:30] When moving the stage using Python, what is the unit of movement? and will new versions of Python be supported in the nanoScope Python toolbox?

General Questions

• [00:15:20] How much work it is to get machine learning up and going, do you need to be an expert to do it?