Rachel Thomas
@math-rachel.bsky.social
AI researcher going back to school for immunology
fast.ai co-founder, math PhD, data scientist
Writing: https://rachel.fast.ai/
fast.ai co-founder, math PhD, data scientist
Writing: https://rachel.fast.ai/
Promises about what AI can achieve with electronic health records must be tempered with the awareness that the data within is too often biased, incorrect, or missing.
(studies on some of the diagnosis delays, pain mismanagement, and biases that are recorded as fact in medical data) 7/
(studies on some of the diagnosis delays, pain mismanagement, and biases that are recorded as fact in medical data) 7/
January 24, 2025 at 10:25 PM
Promises about what AI can achieve with electronic health records must be tempered with the awareness that the data within is too often biased, incorrect, or missing.
(studies on some of the diagnosis delays, pain mismanagement, and biases that are recorded as fact in medical data) 7/
(studies on some of the diagnosis delays, pain mismanagement, and biases that are recorded as fact in medical data) 7/
Some mistakenly believe AI can easily create magic solutions, without understanding the need for high-quality data.
The success of AlphaFold was made possible by 50 years of prior work gathering protein structures into a rich database (Protein Data Bank launched in 1971) 3/
The success of AlphaFold was made possible by 50 years of prior work gathering protein structures into a rich database (Protein Data Bank launched in 1971) 3/
January 24, 2025 at 10:25 PM
Some mistakenly believe AI can easily create magic solutions, without understanding the need for high-quality data.
The success of AlphaFold was made possible by 50 years of prior work gathering protein structures into a rich database (Protein Data Bank launched in 1971) 3/
The success of AlphaFold was made possible by 50 years of prior work gathering protein structures into a rich database (Protein Data Bank launched in 1971) 3/
In addition to the challenge of gathering more data, another challenge of pathology models is needing to capture both local patterns (that show up in a small tile within a slide) and global patterns across the whole slide. 6/
(Image from Chen, et al, 2020, Hierarchical Image Pyramid Transformer)
(Image from Chen, et al, 2020, Hierarchical Image Pyramid Transformer)
January 16, 2025 at 10:13 PM
In addition to the challenge of gathering more data, another challenge of pathology models is needing to capture both local patterns (that show up in a small tile within a slide) and global patterns across the whole slide. 6/
(Image from Chen, et al, 2020, Hierarchical Image Pyramid Transformer)
(Image from Chen, et al, 2020, Hierarchical Image Pyramid Transformer)
The powerful idea behind *foundation models* is to train a on many datasets (e.g. tissue images from many organs) and on multiple tasks (e.g. recognizing cancer, segmenting cells, predicting treatment outcomes)
Patterns learned from one dataset or one task are likely to generalize to others. 2/
Patterns learned from one dataset or one task are likely to generalize to others. 2/
January 16, 2025 at 10:13 PM
The powerful idea behind *foundation models* is to train a on many datasets (e.g. tissue images from many organs) and on multiple tasks (e.g. recognizing cancer, segmenting cells, predicting treatment outcomes)
Patterns learned from one dataset or one task are likely to generalize to others. 2/
Patterns learned from one dataset or one task are likely to generalize to others. 2/
An unmet need in lung cancer research: how to integrate -omics to understand extracellular matrix (ECM) remodeling
(This is the first talk I've seen incorporating the ECM with omics-- it's an interesting perspective!)
Amelia Parker 6/
(This is the first talk I've seen incorporating the ECM with omics-- it's an interesting perspective!)
Amelia Parker 6/
December 3, 2024 at 4:07 AM
An unmet need in lung cancer research: how to integrate -omics to understand extracellular matrix (ECM) remodeling
(This is the first talk I've seen incorporating the ECM with omics-- it's an interesting perspective!)
Amelia Parker 6/
(This is the first talk I've seen incorporating the ECM with omics-- it's an interesting perspective!)
Amelia Parker 6/
The extracellular matrix is a collection of proteins changing over time and space. It has different profiles for different cancer subtypes & profiles.
-- Amelia Parker #multiomics2024 /5
-- Amelia Parker #multiomics2024 /5
December 3, 2024 at 3:58 AM
The extracellular matrix is a collection of proteins changing over time and space. It has different profiles for different cancer subtypes & profiles.
-- Amelia Parker #multiomics2024 /5
-- Amelia Parker #multiomics2024 /5
I can't share videos with 🦋, but there were some neat videos of 3D spatial information from various cancers & the additional info 3D imaging can provide.
Zoe West 4/
Zoe West 4/
December 3, 2024 at 3:50 AM
I can't share videos with 🦋, but there were some neat videos of 3D spatial information from various cancers & the additional info 3D imaging can provide.
Zoe West 4/
Zoe West 4/
Altered cellular metabolism is one of the hallmark features of cancer. This includes altered:
- glycolysis
- oxidative stress
- fatty acid
- amino acid
Spatial data can be used to understand these altered pathways in treatment resistant vs responsive cancer patients
-- Naomi Berrell 2/
- glycolysis
- oxidative stress
- fatty acid
- amino acid
Spatial data can be used to understand these altered pathways in treatment resistant vs responsive cancer patients
-- Naomi Berrell 2/
December 3, 2024 at 3:31 AM
Altered cellular metabolism is one of the hallmark features of cancer. This includes altered:
- glycolysis
- oxidative stress
- fatty acid
- amino acid
Spatial data can be used to understand these altered pathways in treatment resistant vs responsive cancer patients
-- Naomi Berrell 2/
- glycolysis
- oxidative stress
- fatty acid
- amino acid
Spatial data can be used to understand these altered pathways in treatment resistant vs responsive cancer patients
-- Naomi Berrell 2/
Single cell profiling obscures profound spatial heterogeneity.
Even if these 4 samples had the same proportions of cell types, the spatial arrangement is vastly different.
@lochlanfennell.bsky.social #multiomics2024 1/
Even if these 4 samples had the same proportions of cell types, the spatial arrangement is vastly different.
@lochlanfennell.bsky.social #multiomics2024 1/
December 3, 2024 at 2:32 AM
Single cell profiling obscures profound spatial heterogeneity.
Even if these 4 samples had the same proportions of cell types, the spatial arrangement is vastly different.
@lochlanfennell.bsky.social #multiomics2024 1/
Even if these 4 samples had the same proportions of cell types, the spatial arrangement is vastly different.
@lochlanfennell.bsky.social #multiomics2024 1/
CODEX is a multiplexed proteomic technology that lets you stain with 50-120 antibodies.
Yuqi Tan developed the SPACEc python library to provide a streamlined, integrated tool for image extraction, cell segmentation, data preprocessing, & spatial analysis
www.biorxiv.org/content/10.1... 7/
Yuqi Tan developed the SPACEc python library to provide a streamlined, integrated tool for image extraction, cell segmentation, data preprocessing, & spatial analysis
www.biorxiv.org/content/10.1... 7/
December 3, 2024 at 1:45 AM
CODEX is a multiplexed proteomic technology that lets you stain with 50-120 antibodies.
Yuqi Tan developed the SPACEc python library to provide a streamlined, integrated tool for image extraction, cell segmentation, data preprocessing, & spatial analysis
www.biorxiv.org/content/10.1... 7/
Yuqi Tan developed the SPACEc python library to provide a streamlined, integrated tool for image extraction, cell segmentation, data preprocessing, & spatial analysis
www.biorxiv.org/content/10.1... 7/
Biology is *spatiotemporal*
Processes such as cancer, wound healing, & embryonic development occur in both time AND space.
@shazanfar.bsky.social 5/
Processes such as cancer, wound healing, & embryonic development occur in both time AND space.
@shazanfar.bsky.social 5/
December 3, 2024 at 1:21 AM
Biology is *spatiotemporal*
Processes such as cancer, wound healing, & embryonic development occur in both time AND space.
@shazanfar.bsky.social 5/
Processes such as cancer, wound healing, & embryonic development occur in both time AND space.
@shazanfar.bsky.social 5/
Many sources of variation in spatial -omics:
- Tissue structure / library sizes
- Images captured for each FOV (Field of View) separately
- Antibody-binding affinity differences
- Cells overlapping in z-axis
- Partial cells captured
- Background intensity
- Instrument noise
@bhuvad.bsky.social 3/
- Tissue structure / library sizes
- Images captured for each FOV (Field of View) separately
- Antibody-binding affinity differences
- Cells overlapping in z-axis
- Partial cells captured
- Background intensity
- Instrument noise
@bhuvad.bsky.social 3/
December 3, 2024 at 1:11 AM
Many sources of variation in spatial -omics:
- Tissue structure / library sizes
- Images captured for each FOV (Field of View) separately
- Antibody-binding affinity differences
- Cells overlapping in z-axis
- Partial cells captured
- Background intensity
- Instrument noise
@bhuvad.bsky.social 3/
- Tissue structure / library sizes
- Images captured for each FOV (Field of View) separately
- Antibody-binding affinity differences
- Cells overlapping in z-axis
- Partial cells captured
- Background intensity
- Instrument noise
@bhuvad.bsky.social 3/
You need to be careful with how you approach library size normalisation in spatial txomics, or what you could end up eliminating organs / meaningful structures.
-- Dharmesh Bhuva 2/
-- Dharmesh Bhuva 2/
December 3, 2024 at 1:00 AM
You need to be careful with how you approach library size normalisation in spatial txomics, or what you could end up eliminating organs / meaningful structures.
-- Dharmesh Bhuva 2/
-- Dharmesh Bhuva 2/
Spatial biomarkers are associated with immunotherapy response & can help treatment decision making
Tumor cell subtyping of non-small cell lung cancer found 5 distinct tumor clusters which differed in outcomes
-- Ettai Markovits of NucleAI kicking off computational bio session #multiomics2024 1/
Tumor cell subtyping of non-small cell lung cancer found 5 distinct tumor clusters which differed in outcomes
-- Ettai Markovits of NucleAI kicking off computational bio session #multiomics2024 1/
December 3, 2024 at 12:57 AM
Spatial biomarkers are associated with immunotherapy response & can help treatment decision making
Tumor cell subtyping of non-small cell lung cancer found 5 distinct tumor clusters which differed in outcomes
-- Ettai Markovits of NucleAI kicking off computational bio session #multiomics2024 1/
Tumor cell subtyping of non-small cell lung cancer found 5 distinct tumor clusters which differed in outcomes
-- Ettai Markovits of NucleAI kicking off computational bio session #multiomics2024 1/
Unwanted variation exists in single cell data from different domains & requires correction
Seurat & Harmony correct data at embedding-level.
Domain-specific methods (log-norm, CLR, dsb, etc) correct data at feature-level
However, all can end up removing biological signals. New proposed approach:
Seurat & Harmony correct data at embedding-level.
Domain-specific methods (log-norm, CLR, dsb, etc) correct data at feature-level
However, all can end up removing biological signals. New proposed approach:
December 1, 2024 at 7:56 AM
Unwanted variation exists in single cell data from different domains & requires correction
Seurat & Harmony correct data at embedding-level.
Domain-specific methods (log-norm, CLR, dsb, etc) correct data at feature-level
However, all can end up removing biological signals. New proposed approach:
Seurat & Harmony correct data at embedding-level.
Domain-specific methods (log-norm, CLR, dsb, etc) correct data at feature-level
However, all can end up removing biological signals. New proposed approach:
There are many neutrophil-derived structures, incl: exosomes, apoptotic bodies, NET cytoplasts, & extracellular vesicles. Conventional techniques catch < 0.1% of sample volume, missing key structural info
@drjasminewilson.bsky.social research on Aspenglow novel 3D imaging to capture this detail 2/
@drjasminewilson.bsky.social research on Aspenglow novel 3D imaging to capture this detail 2/
December 1, 2024 at 6:20 AM
There are many neutrophil-derived structures, incl: exosomes, apoptotic bodies, NET cytoplasts, & extracellular vesicles. Conventional techniques catch < 0.1% of sample volume, missing key structural info
@drjasminewilson.bsky.social research on Aspenglow novel 3D imaging to capture this detail 2/
@drjasminewilson.bsky.social research on Aspenglow novel 3D imaging to capture this detail 2/
Role of viruses in triggering Type 1 Diabetes:
- viruses theorised as trigger of islet autoimmunity
- both cases & controls show peptides unique to them
- differential immune response to same viruses
- Rather than infection with a specific virus, may be triggered by how the immune system responds
- viruses theorised as trigger of islet autoimmunity
- both cases & controls show peptides unique to them
- differential immune response to same viruses
- Rather than infection with a specific virus, may be triggered by how the immune system responds
December 1, 2024 at 5:46 AM
Role of viruses in triggering Type 1 Diabetes:
- viruses theorised as trigger of islet autoimmunity
- both cases & controls show peptides unique to them
- differential immune response to same viruses
- Rather than infection with a specific virus, may be triggered by how the immune system responds
- viruses theorised as trigger of islet autoimmunity
- both cases & controls show peptides unique to them
- differential immune response to same viruses
- Rather than infection with a specific virus, may be triggered by how the immune system responds
Key research question: can we use *circulating proteins* to predict which melanoma patients will respond to immunotherapy?
-- Fei Yang 6/
-- Fei Yang 6/
December 1, 2024 at 5:29 AM
Key research question: can we use *circulating proteins* to predict which melanoma patients will respond to immunotherapy?
-- Fei Yang 6/
-- Fei Yang 6/
It is not enough to study cancer cells in isolation. Tumors are an ecosystem of interacting & inter-dependent cellular communities: "seed vs soil."
#multiomics2024 5/
#multiomics2024 5/
December 1, 2024 at 5:02 AM
It is not enough to study cancer cells in isolation. Tumors are an ecosystem of interacting & inter-dependent cellular communities: "seed vs soil."
#multiomics2024 5/
#multiomics2024 5/
*Tertiary Lymphoid Structures (TLS)* are complex immune hubs found in some solid tumors. They correlate with more functional B & T cells and improved survival.
-- Tullia Bruno 2/
-- Tullia Bruno 2/
December 1, 2024 at 4:26 AM
*Tertiary Lymphoid Structures (TLS)* are complex immune hubs found in some solid tumors. They correlate with more functional B & T cells and improved survival.
-- Tullia Bruno 2/
-- Tullia Bruno 2/
Many cancers have multiple oncogenes. Does it matter if they are in the same cell or not?
-- Anand Jeyesekaran kicking off the Immuno-Oncology session for #multiomics2024 🧵 1/
-- Anand Jeyesekaran kicking off the Immuno-Oncology session for #multiomics2024 🧵 1/
December 1, 2024 at 4:10 AM
Many cancers have multiple oncogenes. Does it matter if they are in the same cell or not?
-- Anand Jeyesekaran kicking off the Immuno-Oncology session for #multiomics2024 🧵 1/
-- Anand Jeyesekaran kicking off the Immuno-Oncology session for #multiomics2024 🧵 1/
An analogy of the differences between bulk RNA-seq, single-cell RNA-seq, and spatial transcriptomics
#multiomics2024 10/
#multiomics2024 10/
December 1, 2024 at 2:20 AM
An analogy of the differences between bulk RNA-seq, single-cell RNA-seq, and spatial transcriptomics
#multiomics2024 10/
#multiomics2024 10/
The standard breast cancer gene panel is weighted towards variants common in white women, not the variants most common in Black women. -- Jasmine Plummer 9/
December 1, 2024 at 2:14 AM
The standard breast cancer gene panel is weighted towards variants common in white women, not the variants most common in Black women. -- Jasmine Plummer 9/
There is a huge disparity in *whose* genes have been sequenced in the rise of genomic data. This inequality impacts research & healthcare. -- Jasmine Plummer 8/
December 1, 2024 at 2:12 AM
There is a huge disparity in *whose* genes have been sequenced in the rise of genomic data. This inequality impacts research & healthcare. -- Jasmine Plummer 8/
Ongoing challenges for medical foundation models:
- Diversity of Data
- Which resolution? Multiscale Training
- Tiles Centric vs. Cell Centric?
- Transformer based vs other architecture?
- distillation of larger models into efficient smaller models?
- Intrinsic Biases
- Model Drift
- Diversity of Data
- Which resolution? Multiscale Training
- Tiles Centric vs. Cell Centric?
- Transformer based vs other architecture?
- distillation of larger models into efficient smaller models?
- Intrinsic Biases
- Model Drift
December 1, 2024 at 2:03 AM
Ongoing challenges for medical foundation models:
- Diversity of Data
- Which resolution? Multiscale Training
- Tiles Centric vs. Cell Centric?
- Transformer based vs other architecture?
- distillation of larger models into efficient smaller models?
- Intrinsic Biases
- Model Drift
- Diversity of Data
- Which resolution? Multiscale Training
- Tiles Centric vs. Cell Centric?
- Transformer based vs other architecture?
- distillation of larger models into efficient smaller models?
- Intrinsic Biases
- Model Drift