deep maldi for oncology biomarkers study
TRANSCRIPT
Innovation with IntegrityInnovation with Integrity
Deep MALDISerum Proteomic ‘Complementary’ Prognostic Testing for Oncology Research
Rohan A. Thakur, Ph.D.Bruker Daltonics, Asia Pacific
For Research Use Only
Special acknowledgement:Dr. H. Roder, Dr. P. Beresford & Dr. J. Gruia-Gray
Innovation with Integrity
Circulating Blood Proteome

Proteins
Prog
nosi
s
Discov
er
y
Dx
Blood Proteome
Research Biomarker Discovery
Complementary Diagnostics
• Profiling abnormal proteins (as a result of aberrant genes) has significant prognostic potential.
• Can identify patient sub-sets, BEFORE initiation of therapy – patients who will benefit and those who won’t.
• Especially useful when receptor mutation or gene copy numbers don’t relate to disease physiology or pathology
For Research Use Only
Has disease-rich information which can be used to benefit therapy
Innovation with Integrity
Advantage of MALDI-TOF
• Laser based, consistent, reproducible, ionization for proteins• SOP Friendly. No dependency on chromatographic
columns (consumables) or mobile phase formulations.
• Ultra-rugged, self-cleaning technology maximizes uptime (easily do 1000 samples a week)• Low cost of ownership. No dependency on
peripherals like auto-samplers,, or chromatography issues such as carry-over.
• Low maintenance. No solvent or gas lines; no peripherals connections or synchronization required.
• Over 2000 units operating globally in clinical microbiology labs and hospitals• Proven technology platform.
• Technician-friendly, bench-top platform• Simple.
.
Proven platform for clinical
microbiology
>2000 units
operating globally
Class IIMedicalDevice
For Research Use Only
Innovation with Integrity
MALDI and Deep Learning
For Research Use Only
A Perfect Match of Technologies
Deep Learning Basics• Large data sets with good quality • Excels in tasks where the basic unit has very little
meaning in itself, but the combination of such units has a useful meaning
• Biodesix’ clinical applications: classifiers developed for Myelodysplastic syndrome (MDS) and immunotherapy
‘A single MALDI spot can be consistently sampled thousands of times….’Biodesix Deep MALDI patent US20130320203A1
Machine
Learning
Data Input Sources
ModelReady for Prognostic
Testing
Facial recognition is a result of Deep learning
Innovation with Integrity
The Diagnostic CortexTM
Deep learning network to design generalizable and performance-tuned tests based on deep data obtained from Bruker MALDI-TOF mass spectrometers
….
Avoid over-fitting by combining multiple base classifiers
(ensemble average)
Feature spaceSubspace ensemble
Many test candidates
Filter mini-classifiers to design goals
Base classifiers/ complex tests
Train Test
Stratified training/test splits (bags)
Abst
ract
ion
Leve
l
BaseClassifieri
For Research Use Only
Biodesix uses the Diagnostic Cortex to design tests such as the one for Myelodysplastic syndrome (MDS) and immunotherapy
Innovation with Integrity
Rapid Prognostic Capability
For Research Use Only
Blood draw
Separate Serum (Plasma)
Dilute plasma 1:10 with deionized water & spot (1uL)Analyze on MALDI-TOF & process with Biodesix software (TAT for Results <72h)
1234
Innovation with Integrity
Unrivaled (MALDI-TOF) Reproducibility NSCLC Profiling via Proprietary Spectral Processing (veristrat®)
• 9 batches of raw MALDI spectra of 100 technical replicates each of a serum sample:• Large intensity variations both
within and across batches
• The same 9 batches of spectra after spectral processing:• Good reproducibility within and
across batches• Median CV of 9.5%
Inte
nsity
, arb
. un
its
m/z
m/z
Inte
nsity
, arb
. un
its
Before Spectral Processing
After Spectral Processing
For Research Use Only
M Duncan et al., Proceedings of the 61st ASMS Conference on Mass Spectrometry and Allied Topics. 2013. MP 181.
2000Shots
Per Spot
Innovation with Integrity
“Deep MALDI”
• A spectrum in the m/z range where standard MALDI shows few, if any, protein peaks
• The same sample and the same m/z range acquired with Deep MALDI technology showing many (lower) abundance protein peaks
Enhanced sensitivity for measuring >1000’s of serum proteins from <10uL
Deep MALDI with deep learning algorithms uncovers information contained in protein peak profiles that can
be used as a prognostic tool in oncology
For Research Use Only
100K shots per spot
Innovation with Integrity
Results
For Research Use Only
Two prognostic applications for MALDI-TOF MS serum proteomic tests for Oncology
Non-small cell lung cancer (NSCLC)
• VeriStrat®, serum-based test stratifies patients with NSCLC into 2 groups, Good or Poor
• Patients with a VeriStrat Poor status have a more aggressive disease and may be recommended for alternative therapy
• Uses 2,000 shots for each of 3 spots and KNN analysis
1 Vansteenkiste JF, et al. Poster at: The American Society of Clinical Oncology (ASCO); 2012.
• Serum-based test separates patients with MDS into 2 groups, Good or Poor, with vastly different prognosis
• If validated, the test can provide complementary information for physicians to consider in addition to existing scoring systems to aid in planning treatment for patients with MDS
• Uses the UltraFlex Deep MALDI and Deep Learning approach (100K shots per sample)
Roder et al. Abstract #5304, AACR. Biodesix, Boulder, CO.
Myelodysplastic syndrome (MDS)
Innovation with Integrity
Conclusion
• Reflex testing for patients with Non-Small Cell Lung Cancer (NSCLC). The Veristrat® test is a LDT.
• BDX004 (part of FOCAL trial) to be submitted for FDA approval.
• Potential applications to prognostic indications for other tumor types (Breast, Ovarian, MDS) or to define a new classifier that stratifies patients who will benefit from immunotherapy
• Deep MALDI-TOF can be easily done on on commercially available platforms• Deep MALDI is significantly different from
SELDI which focused on measuring individual proteins and their amounts in small data sets.
Proven Prognostic Utility of MALDI-TOF for Oncology Research
For Research Use Only
Liquid biopsies for tumor mutation profiling and host
tumor response