SentenceTransformer based on NeuML/pubmedbert-base-embeddings

This is a sentence-transformers model finetuned from NeuML/pubmedbert-base-embeddings on the train_MNR_hnm_scrna dataset. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: NeuML/pubmedbert-base-embeddings
Maximum Sequence Length: 512 tokens
Output Dimensionality: 768 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- train_MNR_hnm_scrna

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("mariakrissmer/pubmedbert_jonatan_100k_50s_20250711")
# Run inference
sentences = [
    'The pattern MALAT1, JUN, HSP90AA1, JUND, TMSB4X, MT-CO1, RPLP1, RPS27, HSP90AB1, UBC, TPT1, EIF1, H3-3B, DNAJB1, RPS12, RPL10, ACTB, RPL28, RPS3, HSPA6, RPS14, RPL13, MT-ATP6, RPS15, RPL30, FOS, JUNB, RPS27A, NFKBIA, ZFP36, RPS15A, RPL37, PTMA, RPS19, RPL34, RPL11, RPS8, RPL32, MT-CYB, RPL3, RPL19, YPEL5, DUSP1, FTH1, RPS28, TMSB10, DNAJA1, RPS13, HSPE1, KLF2 is indicative of gamma-delta T cell differentiation.',
    'This profile resembles gamma-delta T cell cells, based on genes like MALAT1, TMSB4X, ACTB, MT-CO1, RPLP1, RPS27, RPS12, RPL10, RPL13, NKG7, RPL28, RPS15A, RPS14, CD74, RPS3, RPL12, MT-CYB, RPS27A, RPS19, RPS8, RPL32, RPL30, RPLP2, RPL19, FOS, JUN, PTMA, RPS28, PFN1, RPL34, GAPDH, FAU, MT-ATP6, TMSB10, RPL18, HSP90AA1, COTL1, RPS15, SH3BGRL3, RPL11, RPS23, RPL23A, DNAJB1, RPS24, RPL13A, RPL26, DUSP1, RPL36, H3-3B, RPS6.',
    'The pattern MALAT1, RPL10, RPS27, RPL32, RPL34, RPS15A, RPL30, RPS28, RPS4X, RPS12, RPS19, RPLP1, RPL13, RPL19, RPL11, RPS14, MT-ND3, DNAJB1, RPS13, RPL26, TMSB4X, RPL3, MT-CYB, RPLP2, RPL14, ACTB, RPL8, MT-ND1, RPS15, RPL12, RPL18, RPL28, RPL36, TPT1, RPS6, RPS8, MT-CO1, RPS27A, RPS7, TMSB10, RPL37, PTMA, MT-ND2, RPS24, RPS3, FAU, RPS23, ARHGDIB, HSPE1, ZFP36L2 is indicative of regulatory T cell differentiation.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Triplet

Dataset: train_celltype_scrna_MNR
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	1.0

Training Details

Training Dataset

train_MNR_hnm_scrna

Dataset: train_MNR_hnm_scrna
Size: 303,605 training samples
Columns: sentence1, sentence2, and negative

Approximate statistics based on the first 1000 samples:

	sentence1	sentence2	negative
type	string	string	string
details	min: 160 tokens mean: 177.17 tokens max: 207 tokens	min: 160 tokens mean: 177.56 tokens max: 208 tokens	min: 160 tokens mean: 177.05 tokens max: 201 tokens

Samples:

sentence1	sentence2	negative
`Based on the expression of MALAT1, TMSB4X, GNLY, RPS27, NKG7, MT-CO1, RPL10, RPL13A, RPLP2, RPL34, RPS27A, RPLP1, RPS14, RPS28, TMSB10, RPS12, RPL32, RPL13, RPS4X, RPL26, RPL23A, ACTB, RPS19, RPS15A, RPS6, RPL11, RPL37, RPS3, PTMA, RPL3, RPS24, RPL30, RPL28, RPL19, RPS15, RPL36, RPL35, RPS11, RPS8, RPS23, MT-ATP6, RPL18, RPS13, TPT1, GZMA, RPS7, RPL14, RPL37A, RPL12, MT-CYB, this appears to be a natural killer cell cell.`	`Genes like MALAT1, MT-CO1, TMSB4X, RPLP1, RPL10, NKG7, RPS15A, RPS3, RPL13, RPS19, RPL13A, RPL32, RPLP2, RPS6, RPS8, RPL23A, RPL3, RPL19, RPS14, RPS27, RPL34, RPS15, GNLY, TPT1, RPL35, RPS7, GZMA, RPL18, RPS16, RPL11, EIF1, RPS4X, UBB, RPS23, CMC1, ACTB, RPL28, PFN1, RPS13, RPL37, RPS28, KLRB1, RPS27A, CTSW, RPL14, MT-CYB, MT-ATP6, RPS12, RPL36, RPS11 are hallmarks of natural killer cell cells.`	`The expression of MALAT1, GNLY, RPL10, TSC22D3, FOS, TPT1, NFKBIA, RPLP1, TNFAIP3, RPS27, RPS12, RPL13, RPS27A, JUND, RPL28, ZFP36, H3-3B, RPS23, MT-CO1, RPS8, RPL30, VIM, PTMA, RPS19, RPS14, EIF1, JUN, RPL34, SRGN, RPS6, RPS28, RPL19, RPL32, RPS3, IER2, RPL3, BTG1, RPS15A, RPL12, MT2A, RPS24, RPL11, RPL8, IL7R, RPS15, CXCR4, RPL36, RPL37, CD44, RPL18 aligns with a CD16-negative, CD56-bright natural killer cell, human identity.`
`With genes like MALAT1, FTH1, FTL, SAT1, NEAT1, TMSB4X, MT-CO1, RPS19, TPT1, RPL10, RPL13, RPLP1, RPS27, RPL28, RPS8, RPL34, RPS12, RPS24, EIF1, TMSB10, MT-ND3, MT-ATP6, VIM, NAMPT, RPS13, RPS6, RPL32, RPS23, S100A6, RPS27A, RPL30, MT-CYB, RPS16, RPS15, RPL37, S100A4, RPS28, ACTB, RPS14, RPLP2, RPL12, HSP90AA1, RPL13A, RPL11, RPL8, RPL26, S100A11, RPL37A, SRGN, FAU active, this cell is identified as a non-classical monocyte.`	`FTL, FTH1, ACTB, MALAT1, TMSB10, TMSB4X, MT-CO1, RPL10, RPL28, RPS19, S100A4, S100A6, CST3, CD74, RPS12, COTL1, RPLP1, SAT1, FOS, RPS8, PFN1, SH3BGRL3, IFITM3, RPL30, TYROBP, RPL11, RPS24, IFITM2, RPS13, RPL32, FCER1G, RPL13, RPL34, RPS14, RPS23, RPL12, RPL19, VIM, RPS28, TPT1, PTMA, DUSP1, RPS27A, FAU, RPL8, S100A11, RPL37, PSAP, RPS4X, RPS15 reflect the unique expression profile of non-classical monocyte cells.`	`Observed genes (TMSB4X, MALAT1, FTL, FTH1, TMSB10, SPP1, ACTB, CD74, TPT1, RPL10, GAPDH, RPLP1, RPL13, LYZ, RPS12, VIM, RPL28, RPL13A, RPS14, LGALS1, RPS24, RPS19, RPL32, RPL11, CXCL8, RPLP2, S100A4, S100A6, RPS16, RPS6, RPL8, RPS23, CST3, RPL19, RPS27, PTMA, RPL34, SRGN, RPS15A, RPS27A, RPS3, S100A10, RPL12, RPS20, TYROBP, RPS8, RPL30, NFKBIA, RPS15, RPL26) are indicative of classical monocyte cell function.`
`Top genes from expression profile: MALAT1, HSP90AA1, RPLP1, RPL13, RPL10, MT-ND2, RPL34, RPS6, RPL13A, RPS27, RPS19, HSPB1, MT-CYB, RPLP2, RPL32, RPL26, RPS27A, RPS14, RPL14, RPS28, MT-ND3, RPL3, RPS15, RPL8, RPS8, UBB, RPL36, RPL11, MT-ATP6, HSP90AB1, RPS12, RPL37A, TPT1, RPS15A, RPL12, MT-ND1, ACTB, RPS4X, RPL5, TM4SF1, RPL23A, RPL28, RPL30, RPS7, RPL18, RPS24, HSPH1, RPS3, RPS13, RPS20.`	`Typical epithelial cell of urethra markers such as MALAT1, MT2A, MT1X, RPLP1, MT-CO1, RPS6, RPL34, TM4SF1, RPS19, RPL10, RPL13A, RPS8, RPS27, RPS28, RPL13, RPL11, MT-CYB, RPS4X, RPL3, MT-ND2, RPS27A, MT1M, RPS12, S100A11, ACTB, RPL36, RPL32, RPL28, RPL19, RPS14, RPL37A, RPS15A, FTH1, MT-ND3, RPS15, RPL12, RPL26, S100A6, RPS3, RPL8, RPS24, RPS13, RPL35, RPS23, ANXA2, NEAT1, MT-ATP6, RPL23A, ANXA1, RPLP2 are present in this cell.`	`The pattern MALAT1, RORA, CELF2, GFAP, PHACTR1, TTC3, MT-CYB, CACNA2D3, FAT3, AMZ2, MT-CO1, MT-ND3, ATP2B1, TTC28, TRPS1, DCLK1, PDZD2, ENTREP1, DAAM2, NRG1, SIPA1L1, LINC00609, FYN, CD44, MAP4, NLK, TSC22D1, GLIS3, RASD1, PDE10A, FGF1, AAK1, HIVEP3, HP1BP3, VPS13B, ITPKB, MTDH, EPS8, CAMK4, GRAMD2B, RPL8, NLGN4Y, GPRC5B, TRAPPC9, TANC2, BCL2, ARPP21, BNC2, CHD2, ZHX3 is indicative of astrocyte differentiation.`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim"
}

Evaluation Dataset

Unnamed Dataset

Size: 33,734 evaluation samples
Columns: sentence1, sentence2, and negative

Approximate statistics based on the first 1000 samples:

	sentence1	sentence2	negative
type	string	string	string
details	min: 157 tokens mean: 176.98 tokens max: 200 tokens	min: 158 tokens mean: 177.0 tokens max: 201 tokens	min: 157 tokens mean: 177.29 tokens max: 201 tokens

Samples:

sentence1	sentence2	negative
`A transcriptome with MALAT1, RPL13, RPS8, RPLP1, RPL10, RPS14, RPL32, RPS3, RPS4X, TPT1, RPS12, GAPDH, MT-CO1, RPL3, RPS19, ACTB, RPL11, PTMA, RPS6, RPL5, RPL19, RPL8, RPS7, RPL30, RPS23, RPL18, RPL28, RPS24, RPS15, RPS27A, RPS15A, RPL12, RPL34, RPS13, RPL37, RPL14, VIM, TMSB10, TUBA1B, RPL13A, RPL23A, RPS27, RPS16, RPL36, RPL35, NACA, RPS28, HSP90AB1, RPL37A, RPLP2 points toward mesothelial cell identity.`	`The combination of MT-CO1, RPL10, RPS8, RPS12, MALAT1, RPLP1, RPL13, RPL32, RPS14, RPL34, RPS23, RPS3, RPS27A, RPL11, RPL5, RPL28, RPL30, RPS24, TPT1, MT-CYB, RPS4X, ACTB, RPS19, RPS15A, RPS7, RPL12, RPL37, RPL8, RPL18, RPS13, RPL19, RPS27, VIM, PTMA, MT-ATP6, MT-ND3, RPS15, RPS6, RPL14, RPL3, RPS28, RPL36, GAPDH, NACA, RPL37A, FTH1, TMSB4X, TMSB10, RPLP2, H3-3B is characteristic for mesothelial cell cells.`	`MALAT1, RPLP1, RPL13A, RPL10, PTMA, RPS8, RPS19, RPL13, RPS27, TMSB4X, TMSB10, RPS27A, RPS3, RPS23, ACTB, RPS15A, RPS12, MT-CO1, RPL3, RPL32, RPS14, RPL37A, RPL34, RPS24, RPS6, FTH1, RPS28, LGALS1, RPL23A, H4C3, RPS16, RPL11, RPS7, RPLP2, RPS15, RPL30, RPL28, RPL19, TPT1, RPL37, VIM, RPS4X, RPS20, RPL35, RPL26, RPS13, H3-3B, RPS11, RPL18, GAPDH expression places this cell in the mesodermal cell category.`
`Consistent with memory B cell function, genes like MALAT1, CD74, RPS27, RPLP1, RPL32, RPL13, RPS8, RPS12, RPS28, RPS14, RPL37, RPS15A, RPS23, RPS19, RPL34, RPL10, RPS13, RPS3, RPS15, RPL11, RPL30, RPL28, RPL19, RPS27A, ACTB, RPS6, RPS11, RPLP2, RPS4X, TPT1, RPL37A, RPL12, RPL8, RPL18, RPL3, RPL36, RPL35, FAU, RPL13A, RPL23A, RPS7, RPL14, RPS24, TMSB4X, TXNIP, FTL, RPS16, RPL26, TMSB10, SMCHD1 are expressed.`	`The active transcriptional program includes: TMSB4X, CD74, ACTB, RPLP1, RPL28, RPL10, RPS27, RPS12, RPS6, RPS19, RPS24, RPS15A, RPL37, RPL19, RPL32, RPS8, RPL11, RPL13, GAPDH, RPS27A, RPS3, RPS15, RPL36, RPS14, RPS13, RPS23, RPL23A, RPL18, RPL34, RPL30, RPS7, PFN1, RPS28, RPL8, RPS4X, TPT1, FAU, RPL37A, PTMA, RPL5, RPL12, RPL35, RPLP2, FTH1, RPL3, RPL26, TMSB10, ARHGDIB, NACA, RPS16.`	`This cell shows significant expression of: RPL13A, RPL10, MALAT1, RPL13, RPL3, RPLP1, RPS3, RPS6, RPS12, RPS23, RPS14, RPS27A, RPS19, MT-CO1, RPS8, RPL32, RPL12, RPL8, RPS24, RPS15A, RPL11, RPL26, RPL28, RPL19, RPL34, FTL, RPS7, RPS15, RPL5, RPS4X, RPL37A, RPLP2, RPL14, RPL23A, PTMA, RPS28, RPS27, TPT1, RPL35, MT-ND1, RPL18, RPS16, RPL30, TMSB10, GAPDH, RPS20, RPL36, S100A10, RPS13, RPL37.`
`This cell likely originates from the CD8-positive, alpha-beta T cell family, based on expression of MALAT1, TMSB4X, RPS27, RPLP1, RPL13A, RPL10, RPL13, RPL28, RPS12, RPS15A, RPLP2, RPS19, RPS27A, RPL23A, ACTB, TPT1, RPS3, RPS14, RPL34, FTL, RPL19, BTG1, RPS6, RPL32, RPL30, RPL26, CXCR4, RPS24, RPL11, RPS20, TMSB10, RPL3, RPS16, RPS15, RPS23, PTMA, IL32, RPS8, RPL37, EIF1, RPL12, RPS4X, RPS28, FAU, FTH1, RPS7, RPL14, RPL37A, RPL18, RGS1.`	`This cell fits the molecular signature of CD8-positive, alpha-beta T cell, expressing MALAT1, TMSB4X, RPS27, RPL10, FTH1, RPS12, RPL13, RPLP1, TPT1, RPS27A, RPL30, RPS19, RPL28, RPS15A, RPS28, RPL34, RPL32, RPS3, H3-3B, RPL36, RPL11, RPS23, RPS4X, RPL37, RPS14, PTMA, RPL19, RPL26, H1-10, RPL14, RPL18, FAU, RPL3, RPS13, RPS15, RPS24, RPS8, EIF1, RPL12, BTG1, RPL35, RPL8, RPS7, RPL23A, MT-CO1, RPLP2, PABPC1, FTL, GAPDH, GNLY.`	`Based on the expression of RPLP1, RPL10, RPS27A, RPS12, RPS19, RPL32, ACTB, RPL13, MALAT1, MT-CO1, FTH1, RPS27, RPS6, RPS24, RPS23, RPS8, RPL12, RPS3, IL32, RPL30, RPS4X, RPL19, TPT1, PTMA, RPS28, RPL34, GAPDH, RPS14, HSP90AB1, RPL28, MT-ATP6, RPS13, RPL36, RPLP2, RPL18, FTL, RPS15A, RPL8, RPL14, RPL37A, RPL37, MT-CYB, MT-ND1, RPL5, TMSB4X, RPS15, EIF1, RPL3, RPS16, RPS7, this appears to be a activated CD8-positive, alpha-beta T cell cell.`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim"
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 128
num_train_epochs: 5
warmup_steps: 1000
fp16: True

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 128
per_device_eval_batch_size: 8
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 1
eval_accumulation_steps: None
torch_empty_cache_steps: None
learning_rate: 5e-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 5
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.0
warmup_steps: 1000
log_level: passive
log_level_replica: warning
log_on_each_node: True
logging_nan_inf_filter: True
save_safetensors: True
save_on_each_node: False
save_only_model: False
restore_callback_states_from_checkpoint: False
no_cuda: False
use_cpu: False
use_mps_device: False
seed: 42
data_seed: None
jit_mode_eval: False
use_ipex: False
bf16: False
fp16: True
fp16_opt_level: O1
half_precision_backend: auto
bf16_full_eval: False
fp16_full_eval: False
tf32: None
local_rank: 0
ddp_backend: None
tpu_num_cores: None
tpu_metrics_debug: False
debug: []
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_prefetch_factor: None
past_index: -1
disable_tqdm: False
remove_unused_columns: True
label_names: None
load_best_model_at_end: False
ignore_data_skip: False
fsdp: []
fsdp_min_num_params: 0
fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
fsdp_transformer_layer_cls_to_wrap: None
accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch
optim_args: None
adafactor: False
group_by_length: False
length_column_name: length
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
dataloader_pin_memory: True
dataloader_persistent_workers: False
skip_memory_metrics: True
use_legacy_prediction_loop: False
push_to_hub: False
resume_from_checkpoint: None
hub_model_id: None
hub_strategy: every_save
hub_private_repo: None
hub_always_push: False
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_inputs_for_metrics: False
include_for_metrics: []
eval_do_concat_batches: True
fp16_backend: auto
push_to_hub_model_id: None
push_to_hub_organization: None
mp_parameters:
auto_find_batch_size: False
full_determinism: False
torchdynamo: None
ray_scope: last
ddp_timeout: 1800
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
dispatch_batches: None
split_batches: None
include_tokens_per_second: False
include_num_input_tokens_seen: False
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
eval_on_start: False
use_liger_kernel: False
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional

Training Logs