際際滷

際際滷Share a Scribd company logo

April retreat wet lab

Download as PPTX, PDF
Anthony LaMarca
Anthony LaMarca

The document discusses the development of an automated or semi-automated "smart wet lab" called BIOTURK. It aims to formalize and digitize laboratory protocols and techniques to improve reproducibility, efficiency and knowledge sharing. Version 1 will capture experiment data using cameras and sensors. Version 2 will add analysis of experiments to identify errors or unsafe practices. Version 3 will provide real-time feedback and tutoring to users as they perform experiments. The goal is to develop tools that can help standardize, review and teach complex multi-step biological laboratory procedures.

1 of 15
Download to read offline
BIOTURK THE SMART PROGRAMMABLE WETLAB Klavins/LaMarca ISTC-PC retreat 4/13
Klavins Lab Biochemical Circuit Design / Genetic Engineering / Synthetic Biology What we want to do: Design and implement new genetic programs. What we spend most of our time doing: Transferring, heating, cooling, filtering, and spinning colorless liquids. Refining and optimizing biochemical protocols. DO NOT DISTRIBUTE
Where is the Knowledge? Postdocs brains Tribal Knowledge Superstitions Best Known Methods Electronic notebooks continually eschewed Starbucks Barista NIH Postdoc Starbucks Manager CIF Fellows Postdoc Google Programmer Assistant Professor Position 20 40 60 80 100 Salary in $1,000s DO NOT DISTRIBUTE
Irreproducible Results A study in 2012 by researchers at Amgen showed that only 6 of 50 supposedly seminal results in published cancer research were actually reproducible. Is this what the $30B NIH budget pays for? From Selective cell death mediated by small conditional RNAs, PNAS, 2010. Now being retracted by the authors who could not reproduce their own results (after trying for two years). DO NOT DISTRIBUTE
Industry Has a Problem Too DO NOT DISTRIBUTE
Formalize How-to Knowledge DO NOT DISTRIBUTE
BIOTURK ALPHA SEMI AUTOMATED HUMAN IN THE LOOP WETLAB DONOTCROSSSABSTRACTIONBARRIERDONOTCROSSABSTRACTIONBARRIERDONOTCROSS DO NOT DISTRIBUTE
BIOTURK ALPHA DONOTCROSSSABSTRACTIONBARRIERDONOTCROSSABSTRACTIONBARRIERDONOTCROSS DO NOT DISTRIBUTE
Outcomes New bar for reproducibility Fractal notebooks remember every detail! Diffs on protocols Efficiency Teaching/training Preservation / codification of knowledge DO NOT DISTRIBUTE
-10- New Capstone for 2013: The Smart Wet Lab 10 Similar in many ways to smart kitchen: Fixed space, smaller than a room Tasks performed primarily at a surface Users execute complex multi-step tasks Large number of objects, materials and user actions Can be viewed as a single user or multi-user problem Ripe for machine assistance
-11- Smart Lab v1: Method Capture, Index and Replay 11 Our version 1 smart wet lab will capture: Depth camera video of the lab bench from 2+ angles Possible audio annotation from the experimenter Presence and usage people, tools, containers and other materials Data from instrumented lab tools (e.g.: Pipette that transmits its usage) We will develop a replay tool to review captured data View the audio/video data with synchronized metadata Metadata collected from tags, sensors and camera-based recognition Allows rich indexing and annotation: Queue up Lisas experiment from yesterday Show me when the promethean bromide was being used
-12- Smart Lab v2: Offline Experimental Analysis 12 To object models needed for v1, add The order of steps to be taken and any important timing information A database of the basic lab activities (opening, closing, pouring, shaking, interacting with all the tool and all the machines, etc.) A database of proper practices DOs and DONTs for web labs. e.g.: Do not invert the pipette or let the tip touch the bench Using these inputs, analyze captured data and extend metadata Look for errors in experiment execution (verify materials, actions and ordering) Identify instances in which lab best practice has not been followed Incorporate analysis into the offline experiment viewer: e.g: Show me misuses of the pipette on Tuesday.
-13- Smart Lab v3: Real-time tutoring and Diagnosis 13 Make sense and perception real-time and interactive: The user may announce the intended experiment The system will follow along, in real time offering feedback For senior experimenters: notifications of errors For junior scientists: may make more of a tutorial form Feedback could use both audio and video cues
-14- Smart Wet Lab v1 ISTC for Pervasive Computing 14 Connected lab equipment(scales, pipettes, autoclaves, etc) Lab Bench Structured light Depth camera Microphones Linux PC Windows PC Rfid Reader Bioturk server Web server + Bioturk scripts, protocols and usage data SmartWetLab server Object and activity models + control server + user trace data Touchscreen Tablet Shows Bioturk protocols OpenNI2 <RFID SDK?> bluetooth UHF RFID antennas Perception SDK USB or analog Http (control)NFS (data) http (control) http Actions: - Edit bioturk protocol - View bioturk results - Visualize lab activity data - Playback experiments http NFS http or serial likely (Alternative is a single joint server)
-15- ISTC Ingredients 15 RGBD recognition of lab equipment and actions Sensors/tags that can identify and monitor usage of lab equipment Accurate object localization/tracking via vision or sensors Multimodal recognition algorithms (vision + sensor tag data) Algs to match objects and action recognition to an experiment definition Grounding speech in the larger experiment to allow capture of parametric variations Most of what were developing in the ISTC can be used: RGBD+egocentric camera algs, WISP, IMS, GMTK, grounded speech, etc

More Related Content

April retreat wet lab

  • 1. BIOTURK THE SMART PROGRAMMABLE WETLAB Klavins/LaMarca ISTC-PC retreat 4/13
  • 2. Klavins Lab Biochemical Circuit Design / Genetic Engineering / Synthetic Biology What we want to do: Design and implement new genetic programs. What we spend most of our time doing: Transferring, heating, cooling, filtering, and spinning colorless liquids. Refining and optimizing biochemical protocols. DO NOT DISTRIBUTE
  • 3. Where is the Knowledge? Postdocs brains Tribal Knowledge Superstitions Best Known Methods Electronic notebooks continually eschewed Starbucks Barista NIH Postdoc Starbucks Manager CIF Fellows Postdoc Google Programmer Assistant Professor Position 20 40 60 80 100 Salary in $1,000s DO NOT DISTRIBUTE
  • 4. Irreproducible Results A study in 2012 by researchers at Amgen showed that only 6 of 50 supposedly seminal results in published cancer research were actually reproducible. Is this what the $30B NIH budget pays for? From Selective cell death mediated by small conditional RNAs, PNAS, 2010. Now being retracted by the authors who could not reproduce their own results (after trying for two years). DO NOT DISTRIBUTE
  • 5. Industry Has a Problem Too DO NOT DISTRIBUTE
  • 7. BIOTURK ALPHA SEMI AUTOMATED HUMAN IN THE LOOP WETLAB DONOTCROSSSABSTRACTIONBARRIERDONOTCROSSABSTRACTIONBARRIERDONOTCROSS DO NOT DISTRIBUTE
  • 9. Outcomes New bar for reproducibility Fractal notebooks remember every detail! Diffs on protocols Efficiency Teaching/training Preservation / codification of knowledge DO NOT DISTRIBUTE
  • 10. -10- New Capstone for 2013: The Smart Wet Lab 10 Similar in many ways to smart kitchen: Fixed space, smaller than a room Tasks performed primarily at a surface Users execute complex multi-step tasks Large number of objects, materials and user actions Can be viewed as a single user or multi-user problem Ripe for machine assistance
  • 11. -11- Smart Lab v1: Method Capture, Index and Replay 11 Our version 1 smart wet lab will capture: Depth camera video of the lab bench from 2+ angles Possible audio annotation from the experimenter Presence and usage people, tools, containers and other materials Data from instrumented lab tools (e.g.: Pipette that transmits its usage) We will develop a replay tool to review captured data View the audio/video data with synchronized metadata Metadata collected from tags, sensors and camera-based recognition Allows rich indexing and annotation: Queue up Lisas experiment from yesterday Show me when the promethean bromide was being used
  • 12. -12- Smart Lab v2: Offline Experimental Analysis 12 To object models needed for v1, add The order of steps to be taken and any important timing information A database of the basic lab activities (opening, closing, pouring, shaking, interacting with all the tool and all the machines, etc.) A database of proper practices DOs and DONTs for web labs. e.g.: Do not invert the pipette or let the tip touch the bench Using these inputs, analyze captured data and extend metadata Look for errors in experiment execution (verify materials, actions and ordering) Identify instances in which lab best practice has not been followed Incorporate analysis into the offline experiment viewer: e.g: Show me misuses of the pipette on Tuesday.
  • 13. -13- Smart Lab v3: Real-time tutoring and Diagnosis 13 Make sense and perception real-time and interactive: The user may announce the intended experiment The system will follow along, in real time offering feedback For senior experimenters: notifications of errors For junior scientists: may make more of a tutorial form Feedback could use both audio and video cues
  • 14. -14- Smart Wet Lab v1 ISTC for Pervasive Computing 14 Connected lab equipment(scales, pipettes, autoclaves, etc) Lab Bench Structured light Depth camera Microphones Linux PC Windows PC Rfid Reader Bioturk server Web server + Bioturk scripts, protocols and usage data SmartWetLab server Object and activity models + control server + user trace data Touchscreen Tablet Shows Bioturk protocols OpenNI2 <RFID SDK?> bluetooth UHF RFID antennas Perception SDK USB or analog Http (control)NFS (data) http (control) http Actions: - Edit bioturk protocol - View bioturk results - Visualize lab activity data - Playback experiments http NFS http or serial likely (Alternative is a single joint server)
  • 15. -15- ISTC Ingredients 15 RGBD recognition of lab equipment and actions Sensors/tags that can identify and monitor usage of lab equipment Accurate object localization/tracking via vision or sensors Multimodal recognition algorithms (vision + sensor tag data) Algs to match objects and action recognition to an experiment definition Grounding speech in the larger experiment to allow capture of parametric variations Most of what were developing in the ISTC can be used: RGBD+egocentric camera algs, WISP, IMS, GMTK, grounded speech, etc